ICOS 2010

2010 IEEE Conference on Open Systems (ICOS 2010), December 5‐7, 2010, Kuala Lumpur, Malaysia

ICOS 2010

2010 IEEE CONFERENCE ON OPEN SYSTEMS

Seri Pacific Hotel, Kuala Lumpur 5-7 December 2010 i


CONTENT CONTENT

ii

INTERNATIONAL ADVISORY/LIAISON

iii

REVIEWERS

iii

TECHNICAL PROGRAM

v

J11: Information Management & Education

v

J12: Information Management & Education (cont)

v

J21: Software & Computer Applications

vi

vi

J22: Software & Computer Applications (cont) AUTHOR INDEX

vii

MANUSCRIPT

1

ii


INTERNATIONAL ADVISORY/LIAISON Zaiki Awang, UiTM, Malaysia Chua Tat Seng, NUS, Singapore

REVIEWERS Issac Niwas S Jaime Calvo‐Gallego Jie Wang Joaquín Olivares Bueno K. Y. Sim Kamal Zuhairi Zamli Kashif Saleem Krishna Battula Kriyang Shah Lalji Prasad Li Minn Ang Liam Mayron Lifford McLauchlan Lochandaka Ranathunga M. Emre Celebi M. K. Sheeja Marc Cheong Martin Kuehnhausen Massila Kamalrudin Massudi Mahmuddin Md. Islam Md. Rajibul Islam Mehdi Bahrami Michael Adeyeye Michele Fiorini Mohamad Farhan Mohamad Mohsin Mohamed Rawidean Mohd Kassim Mohammad Hadi Valipour Mohd Helmy Abd Wahab Mouhib Alnoukari Muaz Niazi Muhammad Adib Haron Muhammad Shafie Abd Latiff Mujdat Soyturk Munawar Riyadi

A. Halim Zaim Achilles Kameas Adil Kenzi Akbar Nabiollahi Alireza Ahrary Amad Mourad Angelos Amanatiadis Aniello Castiglione Ankit Salgia Annie Joseph Antti Lahtela Arash Habibi Lashkari Ariffin Abdul Mutalib Arun P v Arun Saha Ashraf Bany Mohammed Azian Azamimi Abdullah Biju Issac Chandra Mohan B Chin Kim On Chitti Babu B Christos Chrysoulas Chung‐Hua Chu Chutisant Kerdvibulvech Daniel Benevides Da Costa Dinesh Sathyamoorthy Eduardo Pinheiro Elisha Nyamasvisva Girija Chetty Hamidah Ibrahim Hasiah Mohamed @ Omar Herdawatie Abdul Kadir Hesam Yousefian Homa Edalati Humaira Nisar Idris El‐Feghi Ihsan Yassin iii


Sami Habib Samraj Andrews Shashikant Sadistap Shivanand Handigund Shohel Sayeed Su Wei Tan Sudarshan Deshmukh Sude Tavassoli Suhas Manangi Syed Khaleel Ahmed Tareq Alhmiedat Tarif Rahman Tee Tiong Tay Tsung Teng Chen Tutut Herawan Velmurugan Ayyadurai Visvasuresh Victor Govindaswamy Vladimir Jotsov Wan Abdul Rahim Wan Mohd Isa Wan Hussain Wan Ishak Wookwon Lee Xiang Cao Zulkarnay Zakaria

Mustafa Man Mustafar Kamal Hamzah Muthukkaruppan Annamalai Nader Anani Naim Nani Fadzlina Natarajan Sriraam Navneet Tiwari Tiwari Ng Keng Hoong Nisar Ahmed Nooritawati Md Tahir Nor Fazlida Mohd. Sani Mohd. Sani Noraziah Ahmad Nordin Abu Bakar Nursuriati Jamil Omari Abdallah Paolo Romano Parminder Reel Paul Soule Rajendrakumar Patil Rajesh Kumar Rajesh Sudarsan Rana Shahid Manzoor Sabu M Thampi Saiful Suliman Salvatore Pileggi iv


TECHNICAL PROGRAM Monday, December 6

J11: Information Management & Education Room: KENANGA Chairs: Letchumanan Chockalingam (Multimedia University, Malaysia), Mohamed Rawidean Mohd Kassim (MIMOS, Malaysia) 2:40 Semantic Query for Quran Documents Results Mohd Amin Mohd Yunus (University of Malaya, Malaysia); Roziati Zainuddin (, Malaysia); Noorhidawati Abdullah (University of Malaya, Malaysia) 3:00 Hard and Soft Updating Centroids for Clustering Y‐Short Tandem Repeats (Y‐STR) Data Ali Seman (Universiti Teknologi MARA, Malaysia); Zainab Abu Bakar (Universiti Teknologi MARA, Malaysia); Noorizam Daud (Universiti Teknologi MARA (UiTM), Malaysia) 3:20 E‐Procurement Current and Future Readiness Level in Malaysia Naseebullah Langove (Universiti Technologi PETRONAS, Malaysia); Dhanapal Dominic (Universiti Technologi PETRONAS, Malaysia) 3:40 Structural Similarity of Business Process Variants Noor Mazlina Mahmod (UTeM, Malaysia)

J12: Information Management & Education (cont) Room: KENANGA Chairs: Letchumanan Chockalingam (Multimedia University, Malaysia), Mohamed Rawidean Mohd Kassim (MIMOS, Malaysia) 4:20 Identification of Magnetizing Inrush Current in Power Transformers Using GSA Trained ANN for Educational Purposes Mehran Taghipour (University of Birjand, Iran); Alireza Moradi (University of Birjand, Iran); Mohammad Yazdani‐Asrami (Babol University of Technology, Iran) 4:40 Making DC‐DC Converters Easy to Understand for Undergraduate Students Reza Ahmadi Kordekheili (Babol University of Technology, Iran); Mohammad Yazdani‐Asrami (Babol University of Technology, Iran); Amir Sayidi (Science and Research branch of Islamic Azad University, Iran) 5:00 Survnvote: A Free Web Based Audience Response System to Support Interactivity in the Classroom Teddy Mantoro (International Islamic University Malaysia, Malaysia); Media A Ayu (International Islamic University Malaysia, Malaysia) 5:20 Semantic Query with Stemmer for Quran Documents Results Mohd Amin Mohd Yunus (University of Malaya, Malaysia); Roziati Zainuddin (, Malaysia); Noorhidawati Abdullah (University of Malaya, Malaysia) 5:40 A Pilot Study in Using Web 2.0 to Aid Academic Writing Skills Azamjon Tulaboev (Universiti Teknologi Petronas, Malaysia); Alan Oxley (Universiti Teknologi PETRONAS, Malaysia) 6:00 Learning Acids and Bases Through Inquiry Based Website Noor Dayana Abd Halim (Universiti Teknologi Malaysia, Malaysia); Mohamad Bilal Ali (Universiti Teknologi Malaysia, Malaysia); Juhazren Junaidi (Universiti Teknologi Malaysia, Malaysia); Noraffandy Yahaya (Universiti Teknologi Malaysia, Malaysia) v


Tuesday, December 7

J21: Software & Computer Applications Room: KENANGA Chairs: Letchumanan Chockalingam (Multimedia University, Malaysia), Mohamed Rawidean Mohd Kassim (MIMOS, Malaysia) 8:40 Adaptive Window Method of FIR Filter Design Abdullah Awad (College of Computer, Anbar University .Iraq, Iraq) 9:00 Using Genetic Algorithm To Break A Mono‐ Alphabetic Substitution Cipher Safaa Omran (Foundation of Technical Education, Iraq); Ali Al‐Khalid (Foundation of Technical Education, Iraq); Dalal Al‐Saady (Foundation of Technical Education, Iraq) 9:20 Single‐Link Flexible Manipulator Control Using Improved Bacterial Foraging Algorithm Heru Supriyono (The University of Sheffield, United Kingdom); M. Osman Tokhi (University of Sheffield, United Kingdom) 9:40 Grid Workflow Recovery as Dynamic Constraint Satisfaction Problem Stanimir Dragiev (Technische Universitaet Berlin, Germany); Joerg Schneider (Technische Universitaet Berlin, Germany) 10:00 Speckle Filtering Of Ultrasound B‐Scan Images ‐ A Comparative Study Between Spatial And Diffusion Filters Sivakumar Ramamurthy (University of Madras, India); Gayathri Kanagaraj Manickam (University of Madras, India); Damodaran Nedumaran (University Of Madras, India) 10:20 Development of Dashboard for Hospital Logistics Management Mahendrawathi ER (Institut Teknologi Sepuluh Nopember, Indonesia); Danu Pranantha (Institute of Technology Sepuluh Nopember (ITS), Indonesia)

J22: Software & Computer Applications (cont) Room: KENANGA Chairs: Letchumanan Chockalingam (Multimedia University, Malaysia), Mohamed Rawidean Mohd Kassim (MIMOS, Malaysia) 11:00 Context‐Aware Ubiquitous Musafir Subrahmanya Venkata Radha Krishna G Rao (Cognizant Technology Solutions, India); Sundararaman Karthik (Cognizant Technology Solutions, India); Jinka Parthasarathi (Cognizant Technology Solutions, India); Prashant Parekh (Cognizant Technology Solutions, India) 11:20 An Information Provision System Based on a Multi‐Hop RFID Scheme for ITS (Intelligent Transportation System) Hiroaki Togashi (The Graduate University for Advanced Studies, Japan); Shigeki Yamada (National Institute of Informatics, Japan) 11:40 A Semantic Index Structure for Integrating OGC Services in a Spatial Search Engine Javier Márquez (University of Castilla‐La Mancha, Spain); José Eduardo Córcoles (University of Castilla‐La Mancha, Spain); Antonio Quintanilla (University of Castilla La Mancha, Spain) 12:00 Prototype of Semantic Search Engine Using Ontology Ahmad Maziz Esa (Universiti Teknologi Petronas, Malaysia); Shakirah Mohd Taib (Universiti Teknologi Petronas, Malaysia); Hong Nguyen (Universiti Teknologi Petronas, Malaysia) vi


AUTHOR INDEX Author

Session Code

J12 J12

Abdullah Awad

J21

Ahmad Maziz Esa

J22

J12

Alan Oxley

J12

J11

Ali Al‐Khalid

J21

J12

Ali Seman

J11

Naseebullah Langove

J11

Alireza Moradi

J12

J12

Amir Sayidi

J12

Antonio Quintanilla

J22

Noor Dayana Abd Halim Noor Mazlina Mahmod

Azamjon Tulaboev

J12

Noorhidawati Abdullah

J11

Dalal Al‐Saady

J21

J12

Damodaran Nedumaran

J21

Danu Pranantha

J21

Dhanapal Dominic

J11

Gayathri Kanagaraj Manickam

J21

Heru Supriyono

J21

Hiroaki Togashi

J22

Hong Nguyen

J22

Javier MÃ¡rquez

J22

Jinka Parthasarathi

J22

Joerg Schneider

J21

JosÃ© Eduardo CÃ³rcoles

J22

Juhazren Junaidi

J12

M. Tokhi

J21

Mahendrawathi ER

J21

Media Ayu

J12

Mehran Taghipour

J12

vii

Mohamad Bilal Ali Mohammad Yazdani‐ Asrami

J11

Noorizam Daud

J11

Noraffandy Yahaya

J12

Prashant Parekh

J22

Reza Ahmadi Kordekheili

J12

Roziati Zainuddin

J11

J12

Safaa Omran

J21

Shakirah Mohd Taib

J22

Shigeki Yamada

J22

Sivakumar Ramamurthy

J21

Stanimir Dragiev

J21

Subrahmanya Venkata Radha Krishna Rao

J22

Sundararaman Karthik

J22

Teddy Mantoro

J12

Zainab Abu Bakar

J11

2010 IEEE Conference on Open Systems (ICOS 2010), December 5-7, 2010, Kuala Lumpur, Malaysia

Semantic Query for Quran Documents Results Mohd Amin MohdYunus, Roziati Zainuddin and Noorhidawati Abdullah Faculty of Computer Science and Information Technology, University of Malaya, 50603, Kuala Lumpur, Malaysia. [email protected],{ roziati,noorhidawati}@um.edu.my

Communication Technologies (ICTs) is more and more important in the development of the society and, consequently, in the improvement of productive processes. A basic issue for achieving that goal is the appropriate management of the process quality. In this sense, they are interested in researching how Semantic Web technologies may improve the quality of the educative process [28].

Abstract- The query-based on the result is less relevant documents results. In this study, a query has been improved in order to retrieve more relevant documents across language boundaries, a mechanism for query translation with semantic which is applied on as semantic query (SQ). Therefore, this study is conducted with the purposes to investigate semantic approach against the queries and vice versa. Furthermore, it is also conducted to investigate the performance query based on total retrieve and relevant. The retrieval however, included the irrelevant documents because of the translation polysemy. Results from the experiments suggest that semantic approach is most important process in cross language information retrieval (CLIR). It also found that semantic approach contributes to better performance in retrieving more relevant and related Quran document results.

III. BASIC FORMULA APPROACH Let W as total words which consists of word 1 (w1), word 2(w2), word 3 (w3) and the rest words (wn) in the search field. Thus the total words like this following formula

Keywords- semantic; speech; query

n

W  w

I. INTRODUCTION

w 1

Current Information Retrieval systems display generally whole documents as result to meet the query given in the process, and provide relevant results as relevant judgment. For IR systems, ontology [9][1] play a key role for adding a semantic dimension between the user‟s query and the data sources. Some attempts for using ontology in search engines can be found in the literature [3]. One difficulty is to find relevant domain ontology, and combine them [2]. Ontology can be also exploited in result retrieval systems during query analysis [12][13]. In conclusion, the system should be flexible [16] in terms of information delivered, from documents to exact answers. Resources are also distributed and heterogeneous, and results have to be integrated. The use of semantic information like ontology can be successfully to retrieve more relevant documents as a whole result.

(1)

where n is the last number of word and i is the first word. Therefore user can input words as many as they want as long as the total of retrieval results from the input words is not influenced. Regarding the result of the words, let D as total retrieval documents related to each word if in each document for the first word (w1d1, w1d2, w1d3….w1d6236), followed by (w2d1, w2d2, w2d3….w2d6236) and last word should be (wnd1, wnd2, wnd3….wnd6236). Therefore concluded result should be also like

D

t

w d d 1

(2)

II. RELATED WORKS Many semantic researches have been conducted through such empirical experiments from various. Ahn et al. [26] improve information access through the use of semantic annotation utilizing a non-traditional approach. Instead of applying semantic annotations to enhance the internal information access mechanisms, they use them to empower the user of an information access system through an innovative named entitybased user interface – NameSieve. The Semantic Web enables an automated, ontology based information aggregation mechanism [27]. The role played by Information and

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

where wn=W and should be any total depending on how many words to be key- in. The meaning of d is related to the verse or ayat or document which consists of the word given in the query. IV. EXPERIMENT APPROACH Qur‟an documents in these experiments have three languages which are the original and holly Quran in classical Arabic language in text, Malay Quranic documents collection [10]

1

which is used by [25] as a domain in their research and English Quranic documents collection [14]. Each collection has 114 surahs and 6236 documents. Every document has its verse and chapter. All documents are as flat files in UTF-8, ASCII or EBCDIC text and searching process is through pattern matching [6]. This research however generally focus on the testing Malay query words are taken from Fatimah‟s collection as natural language queries [7] and the English as well as Arabic query words are translated from the Malay query words and query no. 27 has been tested and evaluated according to the formula in Table 1. Fatimah has obtained them by considering several guidelines put forward by [18] and [21]. Each query would be classified into keywords and replaced by target language. For example, there is Malay query, so it is called as source language and the target language is English or Arabic. Thus English as an example represents the translated word to retrieve English documents and if the query is English, the target language is Malay. The dictionary lists 1325 Arabic, Malay and English words in different flat files including 36 Malay query words selected. The query translation refers to the same index between Malay natural query language [7] and the translation of the natural query language in English or Arabic. When the keyword is Malay, then reference is to the English word at the same index or when the keyword is English and then reference is to the Malay word at the same index. It is considering word by word in the text files.

Result retrieval

User Interaction User Inquiry Query Process

Meaningful words Semantic dictionary files

Semantic Dictionary Look up

Meaningful synonym words

Retrieving Relevant documents Result

V. SYSTEM PROCEDURE Figure 1. The Workflow of Cross Language Information Retrieval Based on Semantic Query

The query can be categorized into two types which are keyword and queryword. If the query is keywords, the results retrieved according to word by word list of verses/documents result and redundant document names existed if merged. But querywords, retrieved according to the whole words as one at all and only when no redundant or unique document names retrieved rankly. Query translation can replace the origin query into another language of the query. This translation is most important for those languages to investigate those information retrieval results. Then, searching process is done according to the type of process of results required. All documents are formatted in “.txt” file for UTF-8, ASCII or EBCDIC text. For searching process, word by word matching is applied on the process. The matching words refer to the words similarity between query and documents in retrieving process. The query submitted to the system is also represented by translated query that is used to search the related files. All steps of SQ result are illustrated in Figure 1. The SQ however also has been applied to be compared with the similarity query to evaluate their results. The results then show significant value which is the better results as related documents to meet the SQ rather than similarity query.

VI. RESULTS AND DISCUSSION All results of the query(ies) translation are referred to the natural language queries of Malay [7] and then translated into Arabic and English queries in this study. Every query is tested to evaluate each result which is matched with manual result as total relevant documents (TRE) for respective query [7]. The evaluation technique is used for precision and recall results [20]. Table 1 shows the formula to calculate the percentage of precision and recall. The SQT is quite suitable with the dictionary that consists of synonyms words and retrieves better results of the most relevant documents. It leads to help to search more and more documents in other languages. These examples also include semantic words after translating the word and matching the words in every document in collection in order to retrieve the most relevant document required from the query given. It is called as CLIR that focuses on search specific language if given query with the same language. Table 2 is referred from [23] thesis for query number 27 to prove that the semantic query in CLIR is significant to retrieve and provide more and more relevant and

2

related results according to the available words provided in semantic files for three languages. Table 3 refers to the semantic query which is comparison between two testing input. Those are single query and semantic query through the process to display two different results according to specific language. In this context, those experiments are involved three languages which are Arabic, English and Malay. Table 4 refers to the retrieval results with the single query according to each language while table 5 refers to the retrieval results with the semantic query according to each language. When the semantic technique applied on query in Table 4, the results are increasing as shown in Table 5 than Table 4. The significant difference shows the total retrieve (TRT) on Malay result at keyword (K), 333 to be 568, on English result at K and Q, 189 to be 13626 and 187 to be 5522 as well as on Arabic result at Query 431 to be 5274. it means that TRT is increasing after applying semantic method correspond to the total retrieve and relevant (TRT).

English

DESCENDANTS PROPHET RELEVANCE EQUATION CHARACTERISTICS MESSENGER

Arabic

‫عالقة األنبياء االصول نبي‬

TABLE 1 RECALL AND PRECISION FORMULA

TABLE 2 NATURAL LANGUAGE QUERIES

Query No. 27

Malay Perkaitan nabi persamaan keturunan ciri-ciri rasul

English Descendants of the prophet relevance equation characteristics messenger

Arabic ‫عالقة‬ ‫األنبياء‬ ‫االصول‬ ‫نبي‬

Language

Language Malay

Q PERKAITAN NABI PERSAMAAN KETURUNAN CIRICIRI RASUL

SQ PERKAITAN NABI PERSAMAAN KETURUNAN CIRICIRI RASUL ASOSIASI HUBUNG KAIT SIMBIOSIS UTUSAN ALLAH PESURUH ALLAH

3

TABLE 4 QUERY EVALUATION TRE TRR Recall(%)

TRT K

Q

333

322

225

English

189

187

Arabic

6340

431

Malay

TABLE 3 SEMANTIC QUERY EVALUATION

RASUL PERTEPATAN PERSERUPAAN ZURIAT DARAH PIUT CUCU SIFATSIFAT LELAKI UTUSAN ALLAH DESCENDANTS PROPHET RELEVANCE EQUATION CHARACTERISTICS MESSENGER CASSANDRA DRUID ASTROLOGER AUGUR CLAIRVOYANT DAYDREAMER DIVINER DREAMER ENTHUSIAST ESCAPIST FORECASTER ACCEPTED ACCUSTOMED ARRANGED AVERAGE BANAL BESETTING BOURGEOIS BUSINESSLIKE CENTRAL CHRONIC COMMON COMMONPLACE CONFORMABLE ‫عالقة األنبياء االصول نبي‬ ‫رسول شاعز هلهن قائد هلهن‬ ‫السوء نذيز الشؤم نذيز‬

Precision(%)

K

Q

K

Q

K

Q

84

81

37.33

36.00

25.23

25.16

225

28

27

12.44

12.00

14.81

14.44

225

293

40

130.22

17.78

4.62

9.28

TABLE 5 SEMANTIC QUERY EVALUATION TRT TRE TRR Recall(%)

Language K

Q

Malay

568

322

English

13627

Arabic

6340

[7] Precision(%)

K

Q

K

Q

K

Q [8]

225

150

181

66.67

36.00

26.41

25.16

5522

225

563

213

250.22

94.67

4.13

5274

225

293

213

130.22

94.67

4.62

3.86[9] 3.72

A. Fatimah, “A Malay language document retrieval system an experiment approach and analysis,” Tesis Ijazah Doktor Falsafah Universiti Kebangsaan Malaysia, 1995. S. Frintrop, E. Rome, and H. I Christensen, “Computational visual attention systems and their cognitive foundations: A survey,” ACM Trans. Appl. Percept., vol. 7, no. 1, Article 6, 2010. T. Gruber, “Toward principles for the design of ontologies used for knowledge sharing,” International Journal of Human-Computer Studies, special issue on Formal Ontology in Conceptual Analysis and Knowledge Representation. Eds, N. Guarino and R. Poli, 1993.

[10] H. Z. Hamidy and H. S. Fachruddin, Tafsir Quran. Translation. Klang, Klang Book Centre, 1987.

VII. CONCLUSION AND FUTURE WORK

[11] A. Katifori, C. Halatsis, G. Lepouras, C. Vassilakis, and E. Giannopoulou, “Ontology visualization methods,” A survey, ACM Comput. Surv., vol. 39, no. 4, article 10, 2007.

The semantic results have a significant difference compared to single results. It has related documents to each another to be presented in the list of each result. The extension work will be focusing on hybrid semantic query (HSQ) results compared to SQ. It is assumed that HSQ is to be considered to more beneficial relevant retrieval results.

[12] V. Lopez, E. Motta, and V. Uren, “Poweraqua: fishing the semantic web,” Proceedings of the European Semantic Web Conference, ESWC 2006, Montenegro, 2006. [13] V. Lopez, V. Uren, E. Motta, and M.Pasin, “AquaLog: An ontologydriven question answering system for organizational semantic intranets,” Journal of Web Semantics, vol. 5, no. 2, pp. 72-105, Elsevier, 2007.

ACKNOWLEDGMENT

[14] T. A. Muhammad and M. K. Muhammad, “Interpretation of the meaning of the noble Quran,” Dar-us-Salam Publications, 1999. http.//www.amazon.com/Noble-Quran-Interpretation-MeaningsLanguage/dp/996074079X.

This research has been funded by the University of Malaya, under the grant number (PS210/2009B) and full-scholarship from the University of Malaya. Thus, I would like to forward our deepest thanks to Prof. Dr. Roziati Zainuddin and Dr. Noorhidawati Abdullah from the Faculty of Computer Science and Information Technology for their endless assistance, technical advice and co-operation.

[15] K. I Normaly, A. R. Nurazzah, and A. B. Zainab, “Terms visualization for Malay translated Quran documents,” In Proceedings of the International Conference on Electrical Engineering and Informatics, Institut Teknologi Bandung, Indonesia, pp. 17-19, 2007. [16] G. Pasi, “Flexible information retrieval: some research trends,” Mathware and Soft Computing, vol. IX, no. 9-1, 107-121, 2002. [17] H. Ping, X. X. Hua, and C. Ling, “Latent attribute Space Tree classifiers,” Journal of Software, vol. 20, no. 7, July 2009, pp. 1735−1745. Institute of Software, the Chinese Academy of Sciences.

REFERENCES [1]

M. A. Aufaure, B. Le Grand, M. Soto, and N. Bennacer, “Metadata- and ontology- based semantic web mining in web semantics and ontology,” D. Taniar & J. Wenny Rahayu eds., Idea Group Publishing, mars 2006, ISBN: 1591409055, 406 p, chapter 9, pp. 259-296, 2006.

[2]

H. Baazaoui-Zghal, , M. A. Aufaure, and N. Mustapha Ben, “A modeldriven approach of ontological components for on-line semantic web information retrieval,” Journal on Web Engineering, Special Issue on Engineering the Semantic Web, Rinton Press, vol. 6, no. 4, pp. 309-336, 2007.

[3]

H. Baazaoui-Zghal, , M. A. Aufaure, and R. Soussi, “Towards an online semantic information retrieval system based on fuzzy ontologies,” Journal of Digital Information Management, vol. 6, no. 5, pp. 375-385, 2008.

[4]

T. d'Entremont and M. A.Storey, “Using a degree of interest model to facilitate ontology navigation,” vlhcc, pp. 127-131, 2009 IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), 2009.

[5]

T. Edward, “The visual display of quantitative information,” press, Chelshire, CT., 1983.

[6]

J. Elly, “The study of existing malay algorithm performed on words beginning with „D‟,” B.Sc. Thesis, Universiti Teknologi MARA, 2000.

[18] M. Popovic and P. Willett, “The effectiveness of stemming for naturallanguage access to Slovene textual data,” Journal Of The American Society For Information Science, vol. 43, no. 5, pp. 384-390, 1992. [19] L. Z. Qiang, C. H. Wu, X. B. Wen, , L. W. Qian, , W. J. Jia, and L. W. Jie, “A fast algorithm for synthesis of quantum reversible logic circuits,” Jisuanji Xuebao (Chinese Journal of Computers)., vol. 32, no. 7, pp. 1291-1303. July 2009. [20] G. Salton and M.J. Mcgill, “Introduction to modern information retrieval,” New York. Mcgraw-Hill, 1983. [21] G. Salton, “Experiments in automatic thesaurus construction for information retrieval,” In Proceedings Ifip Congress 1971, Ta-2, pp. 4349, 1971. [22] C. Ware and P. Mitchell, “Visualizing graphs in three dimensions,” ACM Trans. Appl. Percpt., vol. 5, no. 1, article 2, January 2008. [23] M. A. M. Yunus, “Short query translation: A dictionary-based approach to cross language information retrieval,” Master of Computer Science, Thesis, Universiti Teknologi MARA, Malaysia, 2008.

graphics

[24] R. J. R. Yusof, R. Zainuddin, M. S. Baba, and M. Z. Yusoff, “Visualization systems supporting the reading of Arabic document for non Arabic speakers,” Information Technology Journal, vol. 8, no. 1, pp. 1627, 2009.

4

[25] A. B. Zainab and A. R. Nurazzah, “Evaluating the effectiveness of thesaurus and stemming methods in retrieving Malay translated Al-Quran documents,” In Proceeding Of 6th International Conference On Asian Digital Libraries, pp. 653-662, 2003, Springer-verlag. [26] J.-w. Ahn, P. Brusilovsky, J. Grady, D. He and R. Florian, "Semantic annotation based exploratory search for information analysts," Information Processing & Management, vol. 46, pp. 383-402, 2010. [27] A. Alam, L. Khan and B. Thuraisingham, "Geospatial Resource Description Framework (GRDF) and security constructs," Computer Standards & Interfaces, vol. In Press, Corrected Proof. [28] D. Castellanos-Nieves, R. González-Martínez, C. Soler-Segovia, M. C. Robles, J. Hernández-Franco, M. P. Prendes-Espinosa and J. T. Fernández-Breis, "Semantic Web-based system for managing the educative curriculum," Procedia - Social and Behavioral Sciences, vol. 2, pp. 521-526, 2010.

5


Hard and Soft Updating Centroids for Clustering YShort Tandem Repeats (Y-STR) Data Ali Seman, 1Zainab Abu Bakar

2

Noorizam Daud

Centre for Computer Science Studies, Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA (UiTM), 40450, Shah Alam, Malaysia [email protected]; [email protected]

Centre for Statistics Studies, Faculty of Computer and Mathematical Sciences, Universiti Teknologi MARA (UiTM), 40450, Shah Alam, Malaysia 2 [email protected]

focuses on the hard centroids used by NFKM [7] and the soft centroids used by k-Populations algorithm (KPOP) [6]. Take note that NFKM algorithm is the latest k-Modes-types algorithm for categorical data and KPOP algorithm is the only algorithm uses the soft centroids. Both algorithms have produced good results on categorical data; even though the comparison between these two algorithms is not available yet.

Abstract - This paper compares hard and soft updating centroids for clustering Y-STR data. The hard centroids represented by New Fuzzy k-Modes clustering algorithm, whereas the soft centroids represented through k-Population algorithm. These two algorithms are experimented through two datasets, Y-STR haplogroups and Y-STR Surnames. The results show that the soft centroid performance is better than the hard centroid for Y-STR data. The soft centroid produces 86.3% of the average clustering accuracy as compared 84.3% of the new fuzzy k-Modes algorithm. However, the overall result shows that the hard updating clustering is better than the soft updating clustering while clustering Y-STR data.

II.

Keywords- Clustering algorithm; hard and soft centroids; YSTR; categorical data

I.

INTRODUCTION

Clustering categorical data have attracted many researchers since Huang [1,2] introduced k-Modes clustering algorithm. This includes a fuzzy clustering approach with the introduction of a fuzzy k-Modes clustering algorithm by Huang and Ng [3]. As a consequence, lot of proposed algorithms have been introduced such as k-Representative [4], k-Histogram [5], k-Population [6] and a newly introduced algorithm, called New Fuzzy k-Modes algorithm (NFKM) [7]. On top of that, the k-Modes-types algorithms have also been seen in many aspects of improvement. For examples: (1) Associating with the optimization method such as Tabu Search [8] and Genetic Algorithm [9] (2) Introducing effective initialization methods such as an iterative initial-points refinement algorithm [10] and farthestpoint heuristic based initialization methods [11] and (3) Modifying the dissimilarity measure with a weighting value such as introduced by He et al. [12], Ng et al. [13], San et al, [4], Kim et al. [6 and Ng and Jing [7]. In a different case, Kim et al. [6] have proposed a new technique for updating centroids that is called the soft centroids. This means, the previous updating centroids are classified as the hard centroids. Thus, this paper aims to evaluate the clustering performances of both, the hard and the soft centroids techniques for clustering Y-STR data. The evaluation

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

THE NEW FUZZY K-MODES AND K-POPULATION

In general, the NFKM algorithm focuses on the improvement of the dissimilarity measure from the original fuzzy k-Modes algorithm. The algorithm associates a weighting value for attributes that are similar and maintains the value of one for dissimilar attributes. However, the weighting value is based on the proposed weighting value that had been introduced by Ng et al [13] earlier. However, in this algorithm, the weighting value imposed as the membership values. The new dissimilarity measure with a weighting value is described as Equation (1), (2), (3) and (4). ∑ , (1) , , , where 1, ,

,

,

, ,

1

| |

,

,

(2)

where is the number of objects in fuzzy sense in the ith cluster such that: | |

∑

(3)

is the number of objects with category of the and , jth attribute in the lth cluster in fuzzy sense such that:

6

∑,

,

(4)

,

∑

where 1, ∞ is a weighting exponent or fuzziness index. The KPOP algorithm is totally different. The main idea is to adapt a concept of population in considering centroids chosen. The uniqueness of this algorithm is to compute and obtain the soft centroids in the updating process. Thus, the algorithm begins initiating confident degree, ω before calculating dissimilarity measure. The distance measure is essentially the key intention in assigning clusters. No exact or hard centroid is assigned explicitly like the NFKM algorithm. Thus, the dissimilarity measure and how is the updating centroids done are described in Equation (5), (6), (7), (8), (9), (10) and (11). ∑

,

,

,

where

,

∑

,

,

(6)

,

and ,

Then,

0,

,

,

,

(7)

,

is a normalization factor such that: ∑

THE HARD AND SOFT CENTROIDS

The terms, hard and soft centroids were coined by Kim et al [6] when describing their technique of updating centroids in the KPOP algorithm. However, Ng and Jing [7] when proposing the NFKM algorithm did not make necessary reference to the KPOP algorithm; even though the algorithm had been published earlier. Presently, no comparison has been made for both updating methods. However, the experimental results for categorical data showed that both algorithms produced good results . For example, a common dataset, known as Soybean dataset, both algorithms obtained 100% of the clustering accurracy. See both papers for details. Kim et al [6] described the previous updating centroids of the fuzzy-type k-Modes algorithms as hard decisions (hard centroids) by using membership values. Kim et al [6] also addressed some disadvantages of the hard centroids: (1) the current centroid does not keep information for the next centroids; (2) The algorithm would fall into a local minimum and (3) A single attribute value, with the highest frequency is not sufficient to represent the whole attribute Al. Thus, the proposed soft centroid considers the entire population of all category values while updating the centroids. The main idea is to associate the confident . Therefore, the degree, ω for each є population of the ith cluster centroid is defined as Equation 12, 13 1nd 14.

where ,

(11)

1, ∞ is a weighting exponent or fuzziness index III.

(5)

,

,

(8) , ,…,

,

….,

(12)

,

where The confident degree,

can be calculated as follows: ,

,

∑

0

where:

,

Then,

, ,

0,

,

,1

є

(13)

subject to

(9)

,

|

1, 0

∑

(14)

Unlike the hard centroid, the updating soft centroid is implicitly done while associating and calculating distance measure as described in Equation (5), (6) and (7). The updating centroid is based on the highest confident degree obtained as in Equation (9), (10) and (11). Thus, the attribute is compared to each domain and if it is similar, it takes the membership value, , , , otherwise the value will be zero.

(10)

,

is a normalization factor such that:

7

IV.

a) Dataset 1: This data set consists of 267 records of YSTR haplogroups obtained and filtered from The Finland DNA Project [17]. The original data were 906 that consisted of 7 groups as the time retrieved. However, the data were filtered to choose only 4 groups, called haplogroups, which consisted of L (92), J (6), N (141), and R (28) respectively. The values in the parenthesis indicate the number of records belong to the particular group. Take note that the data were mainly chosen from the groups that had been confirmed by SNP testing only. b) Dataset 2: This data set consists of 236 records of four Surnames: The Donald Surname (112), The Flannery Surname (64), The Mumma Surname (42) and The William Surname (18). The detail description for each data as follows: (i) The Donald Surname consists of 112 records obtained and filtered from the Clan Donald’s DNA Projects [18]. The original data were 896 records. The modal haplotype for this surname is: 13, 25, 15, 11, 11, 14, 12, 12, 10, 14, 11, 31, 16, 8, 10, 11, 11, 23, 14, 20, 31, 12, 15, 15, 16. (ii) The Flannery Surname consists of 64 records obtained and filtered from the Flannery Clan Y-DNA project [19]. The original data were 896. The modal haplotype for this surname is: 13, 24, 14, 10, 11, 14, 12, 12, 12, 14, 13, 30, 16, 9, 10, 11, 11, 26, 16, 19, 29, 15, 15, 17, 17. (iii)Mumma Surname consists of 42 records obtained and filtered from the Mumma-Moomaw Project [20]. The original data were 78. The modal haplotype for this surname is: 13, 25, 14, 11, 11, 14, 12, 12, 13, 13, 13, 29, 17, 9, 10, 11, 11, 24, 15, 19, 30, 14, 17, 17, 17. (iv)William Surname consists of 18 records obtained and filtered from The Williams DNA Project [21]. The original data were approximately 626 records from 94 groups. However, the data were taken from Group 9 only and filtered for 18 records. The modal haplotype for this surname is: 13, 25, 14, 11, 11, 13, 12, 12, 12, 13, 14, 29, 17, 9, 10, 11, 11, 25, 15, 18, 30, 15, 16, 16, 17.

Y-STR

A. Y-STR as Categorical Data Let X ={X1, X2,…, Xn} be a set of n Y-STR data and A ={A1,A2,…, Am} be a set of markers (attributes) of Y-STR data. We define Aj is the jth marker in the actual STR allele value. We define X is a categorical data if it is treated only as a categorical value. Thus, for X categorical data, each attribute Aj describes a domain values, denoted by DOM(Aj). A domain DOM(Aj) is defined as categorical if it is finite and unordered, e.g., for any a,b є DOM(Aj), either a=b or a ≠ b. Consider the jth attribute values are: Aj ={10, 10, 11, 11, 12, 13, 14}, thus the DOM(Aj)={10,11,12,13,14}. We consider every individual has exactly attribute STR allele values. If the value of an attribute Aj is missing, then we denote the attribute value of Aj by a category є which means empty. Let Xi be individual, represented as [xi,1, xi,2,...,xi,m]. We define Xi = Xk , if xi,j = xk,j for 1≤ j ≤ m, where the relation Xi = Xk does not mean that Xi and Xk are the same individual because there exist the two individuals have equal STR allele values in attributes A1,A2,...,Am. B. Y-STR Data and Its Applications Y-STR means Short Tandem Repeats on the YChromosome. The Y-STR data represents the number of times an STR repeats, called allele value for each marker. If a Y-STR marker, say DYS391, the tandem repeats are: [TCTA] [TCTA] [TCTA] [TCTA] [TCTA] [TCTA] [TCTA] [TCTA], thus the allele value is counted as eight. The distance for a person may differ from other by referring the allele values for each marker. If a person shares the same allele values for each marker he is considered to be descended from the same ancestor from a genealogical perspective. In a broader perspective, for instance in studying human migration patterns, it can be applied to whole haplogroups which includes different geographical area throughout the world. The Y-STR data can be grouped into meaningful groups based on the distance for each STR marker. For genealogical data such as Y-Surname project, the distances are typically based on 0 or 1 or 2 or 3 mismatches, whereas the haplogroups are determined by a method known as SNP analysis. All males in the world can be placed into a system of Y-DNA haplogroups named by the letters A through to T, with further subdivisions using numbers and lower case letters [14]. The haplogroups have been established by Y Chromosome Consortium [15]. V.

Both datasets are filtered to: (1) standardize on similar 25 markers (attributes); (2) For Y-STR Surname dataset, the genetic distance is based on 0 to 5 mismatches only. For better results, the experiments were run about 100 times for each algorithm and it was also done randomly reordering the original record. The experiments were also conducted for each fuzziness index, ranging from 1.1 to 2.0. However, the index that produced the best clustering result was only analyzed for final result.

EXPERIMENTAL SETUP

The experiments were conducted on 2 datasets of Y-STR data. The data can mostly be found in worldfamilies.net [16]. The datasets were retrieved on 7th December 2009. The datasets represent: (1) Y-STR haplogroups applications and (2) Y-STR Surname applications. The details for each dataset are as follows:

VI.

EXPERIMENTAL RESULTS

This section discusses clustering performances for hard and soft updating centroids. Thus, the main discussions are

8

based on the performances of the clustering accuracy, the clustering precision and the clustering recall. In addition, the test statistic, t is also carried out for further comparison. Finally, the time efficiency is also reported. We used the same method as Huang [1,2] used in his experiment, which is a misclassification matrix to analyze the correspondence between clusters and the haplogroups of the instances. Clustering accuracy is defined as Equation (15). ∑

than the hard centroid in term of the average clustering accuracy. On the other hand, the hard centroid produced better results for the minimum and maximum of the clustering accuracy. The minimum and maximum values are 58.8% and 97.38% respectively. It is a bit higher as compared to the minimum and maximum values obtained by the soft centroid. See the detail of the clustering accuracy in Table I. Table II and III give insight values of precision and recall. The precision and recall that are very close to 1 indicate the best matching for each pair of cluster and the corresponding class. Since the dataset 2 produced a 100% clustering accuracy, therefore the precision and recall is definitely 1. For the dataset 1, the values for precision and recall are not significant difference. However, the recall values of minimum and maximum, the hard centroid looks better than the soft centroid. It produced a better value of 0.48 and 0.85 for minimum and maximum recall respectively. Further, the hard centroid also produces 0.043 of standard deviation as compared to the soft centroid which is 0.055. See Table III for details. For further comparison, the test statistics, t is carried out. The results of t-test as follows: • Clustering accuracy: The value of the test statistic is t(198) = -1.753, which has a p-value of 0.081. At a 5 percent level of significance, we conclude that there is no significant difference. • Precision: The value of the test statistic is t(172.1) = 1.004, which has a p-value of 0.317. At a 5 percent level of significance, we conclude that also there is no significant difference. • Recall: The value of the test statistic is t(188.3) = 2.119, which has a p-value of 0.035. At a 5 percent level of significance, we conclude that there is significant difference.

(15)

where k, is the number of clusters, ai is the number of instances occurring in both cluster i and its corresponding haplogroups or surnames and n is the number of instances in the data sets. For precision and recall, the calculation is based on Equation (16) and (17) respectively. ∑

∑

(16)

(17)

where is the number of correctly classified objects; is the number of incorrectly classified objects; is the number of objects in a given class but not in the cluster; n is the number of classes/clusters. Table I shows the average clustering accuracy for both methods, the hard and soft centroids. For Dataset 2, both methods obtained 100% of the clustering accuracy as well as the precision and recall. The results indicate that both centroids, either the hard and soft centoids can exactly cluster the Y-STR Surname. It is due to the characteristics of the dataset is made up of different family groups. The algorithms managed to yield a good clustering result. Hence, both centroid methods can cluster the groups even though the distribution among the groups is uneven. For examples, the group of Donald family Surname is too dominant which consists of 112 members as compared to other groups. The lowest number of family members is 18 that belong to William Surname. Take note that the dataset consists of 4 different family surnames. Unlike the dataset 2, the soft and hard centroids did not produce a 100% clustering accuracy. This dataset represents Y-STR haplogroups that consists of many haplotypes from the different haplogroups. However, the soft centroid represented by KPOP algorithm and the hard centroid represented by NFKM algorithm still manage to produce more than 80% of the average clustering accuracy. The KPOP algorithm obtained a bit higher of 86.26% of the average clustering accuracy, whereas NFKM obtained about 84.31% only. Hence both centroids can cluster the Y-STR haplogroups. However, the soft centroid is found to be better

Take note that for the dataset 1, the fuzziness indexes that produced the best clustering result were 1.2 and 1.1 for NFKM and KPOP respectively. However, for Dataset 2, the index was 1.1 for both algorithms. The bold faced numbers refer to the best clustering result obtained by the particular algorithm Table IV compares the time efficiency for both methods. The result shows that the soft centroid obviously takes longer time as compared to the hard centroid. For examples, the soft centroid recorded the maximum time about 95 – 500 seconds, whereas the hard centroid is merely 0 – 2 seconds. Thus, the average time taken for both methods is significantly difference. Thus, the hard centroid is more time efficient as compared to the soft centroid. See the details in Table IV.

9

TABLE I CLUSTERING ACCURACY FOR BOTH ALGORITHMS Dataset

1

2

Evaluation

NFKM

KPOP

Mean

84.31%

86.26%

Minimum

58.80%

44.94%

Maximum

97.38%

TABLE IV COMPARISON TIME TAKEN IN SECONDS FOR BOTH ALGORITHMS Dataset

Evaluation

NFKM

KPOP

Average

0.10

256.9

Minimum

0.06

22.5

93.25%

Maximum

0.16

461.3

Average

1.19

86.0

Minimum

0.92

73.6

Maximum

1.96

97.7

1

Std Deviation

0.074

0.083

P-value

0.081

0.081

Mean

100%

100%

Minimum

100%

100%

Maximum

100%

100%

VII. CONCLUSIONS

Std Deviation

0.00

0.00

P-value

-

-

The ultimate goal of clustering method is to yield a 100% clustering accuracy as obtained by the dataset 2. Hence, the result of the clustering accuracy is the most important element in evaluating the performance of a particular clustering algorithm. However, based on the experiment, it is not so clear whether the soft and the hard centroids produced a better result. In overall result, we can conclude that the hard updating centroid is better that the soft updating centroid because: (1) the hard produces a significant difference for the recall values, which means it may obtain a good result for other datasets or it can be a target to improve it. (2) the hard centroid is exactly time efficient. On the other hand, the idea of having soft updating centorid for clustering Y-STR data is still a good idea. It could be used for future improvement.

2

TABLE II CLUSTERING PRECISION FOR BOTH ALGORITHMS Dataset

1

2

Evaluation

NFKM

KPOP

Mean

0.73

0.74

Minimum

0.51

0.36

Maximum

0.73

0.74

Std Deviation

0.028

0.042

P-value

0.317

0.317

Mean

1.0

1.0

Minimum

1.0

1.0

ACKNOWLEDGMENT

Maximum

1.0

1.0

Std Deviation

0.0

0.0

P-value

-

-

This research is part of our main research of the DNA kinship analyses funded by Fundamental Grant Research Scheme (FRGS), Ministry of Higher Education of Malaysia (Ref. no. 600-IRDC/ST/FRGS.5/3/1293; Project Code: 211201070005). Firstly, we thank Research Management Institute (RMI), Universiti Teknologi MARA (UiTM) Malaysia for their full support of this research. Secondly, we would like to extend our gratitude to many contributors toward the completion of this paper especially the dedication of our research assistants; Mr. Syahrul, Miss Nurin and Miss Soleha.

TABLE III CLUSTERING RECALL FOR BOTH ALGORITHMS Dataset

1

2

Evaluation

NFKM

KPOP

Mean

0.64

0.63

Minimum

0.48

0.41

Maximum

0.85

0.68

Std Deviation

0.043

0.055

P-value

0.035

0.035

Mean

1.0

1.0

Minimum

1.0

1.0

Maximum

1.0

1.0

Std Deviation

0.0

0.0

P-value

-

-

REFERENCES Z. Huang, “A fast clustering algorithm to cluster very large categorical data sets in data in data mining”, Proceedings of the SIGMOD Workshop on Research Issues on Data Mining and Knowledge Discovery, Department of Computer Science, The University of British Columbia, Canada, pp. 1–8, 1997. [2] Z. Huang, “Extensions to the k-Means Algorithm for Clustering Large Data Sets with Categorical values”, Data Mining and Knowledge Discovery, vol. 2, pp. 283–304, 1998. [3] Z. Huang and M. Ng, “A Fuzzy k-Modes algorithm for clustering categorical data”. IEEE Transactions on Fuzzy Systems. 7(4):446-452, 1999. [1]

10

[4]

[5] [6]

[7]

[8] [9]

[10]

[11] [12]

[13]

[14] [15] [16] [17] [18] [19] [20] [21]

O.M. San, V.N. Huynh and Y. Nakamori, “Al Alternative of the K-Means Algorithm for Clustering Categorical Data”, International Journal of Applied Mathematics and Computer Science, vol. 14, No. 2, pp. 241-247, 2004 Z.Y. He, X. Xu, S. Deng, B. Dong, “K-histograms: an Efficient Clustering Algorithm for Categorical Dataset”. In: ARXIV, September 2005 D. W. Kim, K. Y. Lee, D. Lee, Kwang H. Lee, “ A KPopulations Algorithm for Clustering Categorical Data”, Pattern Recognition, vol. 38, pp. 1131-1134, 2005. M.K. Ng and L. Jing, “A new fuzzy k-modes clustering algorithm for categorical data”, International Journal of Granular Computing, Rough Sets and Intelligent Systems, vol. 1, no. 1, pp. 105-118, 2009. M.K. Ng and J.C. Wong, “Clustering Categorical Data Sets Using Tabu Search Techniques. Pattern Recognition, vol. 35, no. 12, pp. 2783-2790, 2002. G. Gan, Z. Yang, J. Wu, “A Genetic k-Modes Algorithm for Clustering Categorical Data. Proceeding of ADMA’05, pp. 195-202, 2005. Y. Sun, Q. Zhu, Z. Chen, “ An Iterative Initial-Points Refinement Algorithm for Categorical Data Clustering”, Pattern Recognition Latters, vol. 23, pp. 875-884, 2002. Z. He, “Farthest-Point Heuristic based Initialization Methods for K-Modes Clustering”, In: ARXIV, October 2006. Z. He, X. Xu, and S. Deng, “Attribute Value Weighting in k-Modes Clustering”, Computer Science e-Prints: arXiv:cs/0701013v1 [cs.AI], Cornell University Library, Cornell University, Ithaca, NY, USA, http://arxiv.org/abs/cs/0701013v1, pp. 1-15, 2007 M.K. Ng, M.J. Li, J. Z. Huang and Z. He, “On the impact of dissimilarity measure in k-modes clustering algorithm”, IEEE Transactions of Pattern Analysis and Machine Intelligence, vol. 29, no. 3, pp. 503-507, 2007. http://www.isogg.org/tree/ISOGG_YDNATreeTrunk08 .html http://ycc.biosci.arizona.edu/ http://www.worldfamilies.net www.familytreedna.com/public/Finland/default.aspx (Retrieved May 3, 2010) http://dna-project.clan-donald-usa.org. (Retrieved May 3, 2010) http://www.flanneryclan.ie/dnaResults.htm (Retrieved May 3, 2010) http://www.mumma.org/ (Retrieved May 3, 2010) http://williams.genealogy.fm/results_page.php (Retrieved May 3, 2010)

11


E-Procurement Current and Future Readiness Level in Malaysia Naseebullah1, Shuib Bin Basri2 and P.D.D. Dominic3 Department of Computer and Information Sciences Universiti Teknologi PETRONAS Bandar Seri Iskandar, Tronoh Perak, Malaysia [email protected], [email protected], [email protected] where one or more companies try to source their suppliers at lowest costs possible [9]. E-procurement significantly reduces costs, reduces paperwork, lowers administrative costs, brings better quality and improves delivery [10], [11] and [12]. According to a survey data 11% to 12% business growth and 35% cost reductions has been experienced by organizations after E-procurement implementation [13].

Abstract—E-procurement has been recognized as an element of Business-to-Business (B2B) E-commerce. In B2B organizations have inter firm acquisition of goods and services over the internet. This paper focuses the current and future readiness level of E-procurement among organizations in Malaysia. A survey questionnaire was administered to collect data from 46 organizations. The general finding shows that Malaysian medium and large organizations have positive influence towards online procuring in terms of readiness. This paper also provides information about future intentions of Eprocurement implementation.

The objective of this study is to explore the current and future status of E-procurement implementation among Malaysian organizations. As before no such studies has been done about organizations E-procurement readiness level in Manufacturing and Services sector because these two sectors are the back bone of the country economy. As to know the Malaysian firms feedback, we have conducted a personal administrative questionnaire in Central Malaysia and Perak using SPSS for data analysis. The result generally support 46.7% organizations using E-procurement, either some have in house integrated E-procurement system and other organizations are engaged in outsourcing as well as collaborations with trading partners.

Index Terms—Internet, E-procurement readiness level, Malaysia

I. INTRODUCTION

T

he emergence of internet technology significantly changed many firms operations and become a universal source for public, government and business communities [1]. Internet is believed one of the best means for companies to attract customers and gain bigger market [2]. Internet is a powerful business tool, many Asian companies have moved quickly to take advantage of electronic commerce (E-commerce) [3]. Most of the companies developing plans to integrate internet-based ecommerce into their supply chain management practices to maintain a competitive advantage [4]. Specifically, Ecommerce has one of most important supply chain management application is Business-to-Business (B2B) also known as Electronic procurement (E-procurement). Procurement represents one of the largest expense items in a firm’s cost structure and considered as a strategic player in the value chain [5]. In procurement, there are two purchasing categories, Direct procurement (raw material, spare parts for producing finished products) and Indirect procurement (maintenance, repair and operating supplies).

The remainder of this paper is structured into six sections. Related work will be the second section. Research methodology will be introduced in the third section. Research findings and results discussion will be described in the fourth and fifth section respectively, while conclusion and future work will be in the sixth section. II. RELATED WORK According to survey findings of (2004) National eProcurement Research Project Australia (NeRPA) readiness means the current use and prospective levels of adoption of E-procurement by organizations [14]. According to [15] organizational readiness and IT infrastructure degree has been found as a successful predictors of IT adoption. Basically E-procurement involves the use of internet for purchasing goods and services. Eprocurement reduced the all manual work with simple clicks of the mouse [16]. The delivery of order is made possible on the same day by E-procurement use [17]. E-procurement has a lot of benefits in terms of cost reduction, lower managerial complexity and reduces the number of suppliers and also improves the cycle time and purchasing process [1]. Because of these benefits the required product reach quickly in market and helps to capture the market share [10]. E-procurement gives short and long term benefits to the organizations and it assimilates global supply chain and reduces cost and time [18].

E-procurement has different elements, including electronic ordering, internet bidding, purchasing cards, reverse auctions and integrated automatic procurement systems [6]. E-procurement is defined as “the integration, management, automation, optimization and enablement of an organization’s procurement process, using electronic tools and technologies, and web-based applications” [7]. Eprocurement refers to “the electronic acquisition of goods and services in a firm” [8]. E-procurement sites also recognized as (B2B) marketplaces, electronic supply chains, trading communities and web-based procurement networks

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

1 12

III. RESEARCH METHODOLOGY

CEO Director General Manager Manager Officer Others Total

This study investigated the current and future readiness level of E-procurement in Malaysia. Specifically, the survey questions focused the following information: •

•

Demographic profile (in terms of respondent gender, position, education and organization profile in terms of sector, operating business and number of full time employees and procurement staff) Organization E-procurement current and future readiness level in Malaysia (in terms of Internet usage, web portal, procuring online extent and purchase type category)

Table 3: Education Education PhD Master Bachelor Diploma Total

Table 4: Type of Sector Type of Sector Manufacturing Services Others Total

Frequency 22 15 9 46

Percent 47.8 32.6 19.6 100.0

The Table 5 shows organizations operations, either conducting business within the country (means locally) or overseas (that is globally conducting). About 58.7% organizations were found globally operating while 41.3% were engaged in locally operations. Table 5: Conducting Business Conducting Business Frequency Locally 19 Globally 27 Total 46

A. Respondent’s Profile A total 60 organizations were visited in May 2009 and 57 organizations accepted for questionnaire feedback. From 57 there were 46 complete questionnaires received and the findings reported here are based on the analysis of this data.

Percent 41.3 58.7 100.0

Most important feature of the organization is the size of the firm in terms of full time employee numbers. According to the statistics results in Table 6 shows 83.7% organizations fall within the Small Medium Enterprises (SMEs). According to Malaysian SMEs definition 150 full time employees are listed in the SMEs. 17.3% organizations were find more than 150 full time employees. Overall these statistics shows small and large organizations operating locally and globally in Malaysia.

As shown in Table 1, about respondent gender, most of the respondents were male with 69.6% and female with 30.4%. Percent 69.6 30.4 100.0

The Table 2 and 3 shows most important feature of the respondent’s is his/her position and education levels, most of positions were Manager and General Manager 28% and 26.1% respectively. Frequency

Percent 4.3 21.7 56.5 17.4 100.0

In Table 4, the country most two important sectors that are manufacturing and services which play a vital role in the economy. In this data manufacturing and services frequency were recorded 47.8 and 32.6% respectively and 19.6% were others sectors. Although these two sectors covered all type of manufacturing and services sectors and the remaining were count as others.

IV. RESEARCH FINDINGS

Table 2: Position Position

Frequency 2 10 26 8 46

B. Organization’s Profile

The Statistical Package for Social Science (SPSS) has been used to analyze the survey data. For this data interpretation we only use the (Bar and Pie Chart) graphs to analyze the data and just to know the level of organizations.

Frequency 32 14 46

6.5 13.0 26.1 28.3 17.4 8.7 100.0

While in terms of education that Bachelor degree was recorded high (56.5%) then master and diploma with 21.7 and 17.4% respectively. Over all this statistics suggest that the respondents were generally well position and educated.

Our research methods include literature review, questionnaire design, pilot study, survey and data analysis [19]. The above questions and objectives were formulated on existing literature than we design the above question items. A pilot study was conducted to refine the format of the questionnaire with addition and removal as well as with rephrasing of several items. This study employed a cross sectional field study survey method, using self administered questionnaires [18], distributed in Selangor and Perak Malaysia as an initial survey feedback. The remaining states will be covered in near future for completing sample size of the Manufacturing and Services sectors in Malaysia. Sixty organizations were randomly selected from commerce dotcom Business Directory (2009). Subsequently, 60 organizations were visited with the approval letter stating the objective of the study for conducting survey. From 60 organizations, 57 responses were collected, 11 questionnaires were incomplete and rests of 46 responses were usable.

Table 1: Gender Gender Male Female Total

3 6 12 13 8 4 46

Table 6: Number of full time Employees Number of Employees Frequency Less than 10 9 11-20 8 21-30 5 31-40 1

Percent

2 13

Percent 19.6 17.4 10.9 2.2

41-50 51-100 101-150 More than 150 Total

4 5 6 8 46

organizations are SMEs and 19.6% staff were counted (4-6) people. About 10.9% were found in (more than 10) people and these staff usually from large organizations.

8.7 10.9 13.0 17.3 100.0

Figure 3: Procurement Staff 16

C. E-Procurement current and future readiness level

14

In this section, we identified the current and future readiness level of E-procurement among Malaysian organizations. Fig. 1 shows that 95.7% of the companies currently using the internet, so this is one of basic source for information as well as a tool for online trading. It also shows positive intentions towards E-procurement among organizations.

12 10

8

6

Figure 1: Internet Access

4

Count

50 2 0 None

40

1-2

2-4

4-6

6-10

More than 10

Procurement Staff 30

Fig. 4 describes a pie chart data of E-procurement level in four parts, the major one part shows 45.7% organizations already involved in online procuring while 17.4% have currently under consideration to procure online. About 21.7% organizations have intentions to procure online within 1-2 years and rest of 15.2% has no intention to procure online. This statistics suggests a positive influence of organizations towards online procuring, only some of the organizations have no intention because of their business nature, where they don’t need to procure online.

20

Count

10

0 Yes

No

Internet access

Figure 4: E-Procurement Level Fig. 2 indicates 89.1% of the total respondents have their own websites, but no details about using websites. This also shows a good impact of Malaysian organizations for promoting their organizations at global level.

No Intention to proc 15.2%

Figure 2: Firm Web Portal 50

Already Procuring Currently under

45.7%

17.4%

40

30 Intend to procure 21.7%

20

Fig. 5 presents the E-procurement extent level in the organizations; either they have their own integrated Eprocurement system or outsourcing their products through third party. According to fig. 4 statistics results 45.7% organizations are engaged in online procuring, so in fig. 5 the 45.7% organizations E-procurement extent is shown. The major one extent level is 26.1%, where organizations have in house integrated E-procurement system and these are larger and medium organizations operating globally. About 10.9% are engaged in outsourcing, using third party E-procurement like marketplaces. 8.7% have collaboration

Count

10

0 Yes

No

Firm Web Portal

Fig. 3 provides information about the number of procurement employees. About 35.4% employees involved in procurement (1-2), because in this study 83.7%

3 14

with their trading partners. The rest of 54.3% have no yet intention to use E-procurement either outsourcing or collaboration and these organizations are micro and small in terms of SMEs definition.

level in Malaysian base organizations, which is 45.7% have already procuring online and most of these organizations were medium and large, while 17.4% of small and some medium organizations were found currently under consideration of E-procurement implementation. About 21.7% organizations have intentions to procure online within 1-2 years, so results show a positive influence for upcoming years of E-procurement implementation and procuring online. Only 15.2% organizations have no intention to procure online and these were found micro and some small organizations. Basically these organizations have different challenges in terms of firm size, lack of awareness and some stated the reason about the nature of the business where procuring online is not applicable. Beside E-procurement level our research findings also shows the E-procurement extent level, either organizations have their own integrated system or outsourcing through marketplaces or collaboration with trading partners. In this study we tried to focus in house integrated E-procurement system but for general perceptions and intentions we also look over other organizations. In this statistics 26.1% have in house integrated E-procurement system, about 10.9% procure online through outsourcing and 8.7% have collaborations with their trading partners.

Figure 5: E-Procurement Extent

In house intergrated 26.1%

None 54.3%

Outsourcing 10.9%

E-Collaboration 8.7%

Fig. 6 provides information about purchasing type of direct procurement, indirect procurement, services and all these. In this graph the highest purchasing type is direct procurement which support for producing finished products. The second highest bar shows all these (means most of organizations engaged in all purchasing type). The remaining two bars shows indirect procurement and services which helps in operating of supplies.

VI. CONCLUSION AND FUTURE WORK E-procurement is one of the responsive tools for supply chain; due to well integrated supply chain the organizations quickly get the right products and services at their right time. For such situation organizations are trying to integrate their business operations at domestically as well as globally to capture the market and improve supply chain for on time production and delivery of products to their customers. This study basically investigated the current and future readiness of E-procurement in Malaysian base organizations. The statistics result suggests that medium and large organizations have strong influence for online procuring of direct procurement, indirect procurement and services over the internet. Most of the large organizations have in house integrated E-procurement system and small and medium organizations were found in outsourcing of their products through marketplaces and some were found in collaboration with their trading partners. This study is helpful for further investigation in term of challenges and opportunities of E-procurement in Malaysian base organizations than we come up with strong recommendations and strategies for Malaysian base organizations as well as it will be helpful for developing countries.

Figure 6: Purchase Type Categories 30

20

Count

10

0 Direct Material

Indirect Material

Services

All these

Purchasing type

V. DISCUSSION The finding of this study provides a clear picture about the E-procurement current and future readiness level among Malaysian organizations. In this study we focused micro, small, medium and large organizations because we identify the over all intentions and their perceptions about Eprocurement implementation. Through this survey result the usage of the internet among organizations is more than 95% and 89.1% organizations have their own web portal which is a positive influence towards online business. The most two other important figure are E-procurement

REFERENCES

4 15

[1]

Naseebullah, F.B.H. Mohd and PDD. Dominic, “The role of Businessto-Business E-procurement implementation”, Proceeding of 4th International Symposium on Information Technology, Vol. 1, pp. 1-4, 2010, ISBN#: 978-1-4244-6715-0

[2]

Tan, K.S., Chong, S.C. and Lin, B. (2009), “Internet-based ICT adoption among small and medium enterprises: Malaysia’s perspective”, Industrial Management & Da ta Systems, Vol. 109 No. 2, pp. 224-44.

[3]

B.K. Chia and A. Suliman, “The adoption of Electronic Procurement in Singapore”, Electronic Commerce Research, 2:, 61-73, 2002

[4]

T.M. Rajkumar, “E-Procurement: Business and Technical Issues”, Information Systems Management, 18: 4, 1-9, (2001)

[5]

G.A. Mohamed, “Predicting e-procurement adoption in a developing country: An empirical integration of technology acceptance model and theory of planned behaviour”, Industrial Management and Data Systems, Vol. 110 No. 3, pp.392-414, 2010

[6]

M.J. Moon, “E-procurement management in state governments: Diffusion of e-procurement practices and its determinants”, Journal of Public Procurement 5 (1), 54-72, 2005

[7]

V. Tatsis, C. Mena, L.N.V. Wassenhove and L. Whicker, “Eprocurement in the Greek food and drink industry”, Journal of Purchasing and Supply Management, 12, 63-74, 2006

[8]

E. Turban, D. King, D. Viehland and J. Lee, “Electronic Commerce 2006: A Managerial Perspective, Pearson/Prentice-Hall, Englewood Cliffs, NJ, 2006

[9]

D. Ong, “Putting B2B Hype in Perspective”, Business Times (Singapore), 2000

[10] D. Thomson and M. Singh, “An e-procurement model for B2B exchanges and the role of e-markets” Sixth annual collector conference on electronic commerce, Coffs Harbour, Pacific Bay Resort, p. 227–37, 2001 [11] R. Hsiao and T.S.H Teo, “Delivering on the promise of eprocurement” MISQ Executive, 4(3):343–60 2005 [12] P. Hawking, A. Stein, D. Wyld and S. Forster S, “E-Procurement: Is the Ugly Duckling Actually a Swan Down Under”, Asia Pacific Journal of Marketing and Logistics, Vol. 16, No. 1, 1-26, 2004 [13] T.S.H Teo, S. Lin and K.H Lai, “Adopters and non-adopters of eprocurement in Singapore: An empirical study”, Omega 37 (2009), pp. 972 – 987, (2008). [14] S. Williams and K.L. Smith, “National eProcurement Research Project Australia (NeRPA)” survey, (2004). [15] C. Iacovou, I. Benbasat, A. Dexter, Electronic data interchange and small organizations: adoption and impact of technology, MIS Quarterly 19 (4) (1995) 465–485. [16] T. Chien, D. Ahrens, “E-procurement: the future of purchasing”, Circuits Assembly, 12(9): 26-32, (2001). [17] W.D Presutti, “Supply management and e-procurement: creating value added in the supply chain” Industrial marketing management, vol. 32, no. 3, 2003. pp. 219-226. [18] A. Gunasekaran and E.W.T. Ngai, “Adoption of e-procurement in Hong Kong: An empirical research”, International Journal Production Economics 113, pp. 159–175. (2008). [19] L. Ziqi and S. Xinping, “Consumer perceptions of internet-based eretailing: an empirical research in Hong Kong”, International Journal of Services and Marketing, Vol. 23/1, pp. 24-30, (2009)

5 16


Structural Similarity of Business Process Variants Noor Mazlina Mahmod, Wong Yan Chiew Faculty of Electronics and Computer Engineering Universiti Teknikal Malaysia Melaka (UteM) Melaka, Malaysia {mazlina, ycwong}@utem.edu.my

based on the notion of process similarity, where the structural aspects of the process variants are compared according to specific query requirements. The advantage of this approach is the ability to provide a quantitative measure for the similarity between process variants, which further facilitates various BPM activities such as process reuse, analysis and discovery. In the subsequent sections, we first present the preliminary work including motivation and related work for this topic. Then, we discuss the overall variant management framework introduced in [7] and PVR (Process Variant Repository) architecture in [5] that adopted for this research. Later in fourth section, we introduce the notion of structural similarity. This notion is used to carry out the structural similarity analysis and similarity degree computation process, which are respectively presented. The fifth section will discuss the result of our approach. Finally in the last section, we evaluate and conclude this paper with a summary of the work contributions and the noteworthy expansions for future research.

Abstract— Variance in business process can lead to various changes and modifications of business requirements, strategies and functionalities since it is a valuable source of organizational intellectual capital and represents a preferred and successful work practice. It is important to provide an effective method to analyze the similarity between these variants since it can bring benefits for organization productivity. Through this paper, we propose an efficient approach for undertaking the structural similarity analysis and subsequently providing a formula to compute the degree of similarity between the structural relationships of the variants in a systematic way. We hope that the systematic method introduced in this paper can be applied successfully to solve the ambiguity issue in defining and measuring the structural similarity between the process variant. Keywords-Process Variants, Structural Analysis, Similarity Degree

I.

INTRODUCTION

Variance or instance adaptation in business process is resulted from the changes in business strategies, constraints and the emergence of unexpected events such as disconnection between documented models and business operation, flexible and ad-hoc requirements, dynamic change that cannot be handled by exception handling policies, collaborative and/or knowledge intensive work. As a result, the execution of process instances needs to be changed at run-time causing different instances of the same business process to be handled differently according to instance specific conditions. This situation leads to the production of a large number of process variants. These process variants reflect the awareness of process constraints and requirements which provide valuable insight into work practice, help externalize previously tacit knowledge and provide valuable feedback on subsequent process design, improvement and evolution.

II.

A. Motivation Motivated by the currently inadequate generic approach for similarity analysis between process variants and the importance of structural aspect as foundation for every process model, therefore, we are highly interested to find an effective approach for analyzing structural similarity between process variants and introduce a constructive approach to compute the degree of the structural similarity. We believed that our approach introduced in this paper can assist process designers to discern the type and the degree of structural similarity between process variants as to improve their decision in producing a better business process model in the future. B. Related Work There are several related works that contribute to this research. One of them is a notable work for business processes that has been addressed in process equivalence in [2] and another one is an approach about similarity analysis between process variants presented in [5]. The process equivalence described in [2] provides detail information about the execution sequences to conduct the similarity analysis. The approach presented in [5] explains on how to detect semantic business process variants using ontology approach. We also rely on the graph reduction technique proposed in [4] which presents five

It is imperative that well-organized support for managing these process variants be extended to organizations mature in their BPM (Business Process Management) uptake so that they can ensure organization reliability and steadiness, promote reuse and capitalize on their BPM investments. This paper presents an approach for managing business processes that is conducive to dynamic change and the need for flexibility in execution. The approach is based on the notion of process structure constraints. It further provides a technique for effective utilization of the adaptations manifested in process variants. In particular, we will present a facility for discovery of preferred variants through effective search and retrieval

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

MOTIVATION AND RELATED WORK

17

the search and retrieval activities of process variants. Further information about the reference architecture of PVR can be referred in [5].

effective reduction rules (i.e. terminal, sequential, closed, adjacent and overlapped) to verify structural correctness of process variant. For improvement in process structure reduction, we also adopt the selective reduce technique in [6] to modify the graph reduction technique in [4] by eliminating only the task nodes in process variants that are not contained in the node set of process query. This technique allows the process variants to be reduced into a non empty graph that is similar or almost similar with the process query. We applied both graph reduction techniques and algorithms in [4] and [6] to identify which process variants are exactly similar or slightly similar when being compared to a user-defined process query. Besides, we also enhance the flows counting algorithm in [6] to develop and improve our structural similarity degree computation formulas as presented in the fourth section. III.

IV.

STRUCTURAL SIMILARITY ANALYSIS

There are a number of notions and theories available that have been used to define the structural relationship between process variants such as in process equivalence in [1] and process subsumes and transform relations in [4] and [6]. However, the problem about the degree of similarity still remains because it is difficult to measure the degree of slightly similarity between the structural elements and relationships exist in the process variants.

PROCESS VARIANT MANAGEMENT

Process variants are very complex objects as they may vary significantly due to the complexity of data required to describe them even though they satisfy the same set of constraints. As consequence, it is not easy and very complicated to capture the type and compute the degree of structural similarity between them. To treat this problem, we apply Petri nets approach and its tool to model and analyse the structural aspect between the process variants. Petri nets have a mathematical foundation to check the algorithms of process model at various abstractions and can be used as a design language for the specification of complex process structure. Please note that we only make use of the basic terminology of Petri nets and its subclass, Free Choice Petri nets (FCPN) to model the process variants presented in this paper with the purpose of reducing the semantic complexity in modelling the process models. More background material for Petri nets can be referred in [1, 3, 9, 10]. Please see [11] and [12] for detail information about FCPN. Instance Adaptation

5 3

Executed Process Instance

Figure 2. Conceptual Approach of Structural Similarity Analysis

To solve the problem, we will introduce an effective means through our conceptual approach of structural similarity analysis as summarized in Figure 2. Throughout the remaining subsections, we will present and discuss how this approach is used to analyse, measure and compute the structural similarity degree between the process variants in a systematic way.

PVR Usage

A. Process Variants Modeling An example of business process model illustrated in Figure 3 demonstrates the flow of activities in a vendor performance assessment process. Task like registering, assessing performance and selecting a vendor are predictable and repetitive. Somehow, there are a number of assessments tests can be performed based on preferred order during the assessing vendor performance procedure. The tests (i.e. T1, T2, …, Tn) represent the tests that will be executed as assessment test for the vendor. Each process variant prescribed uniquely for each case that has to be coordinated and controlled. We have select a set of test (i.e. T1, T2,…,T6) to represent one of the variant for test assessment during assessing vendor performance process. Then, we restructure and remodelled the example of business process in Figure 3 using Petri nets and FCPN approach as presented in Figure 4. We apply fundamental terminology of Petri nets to define our new model in Figure 4 as follows:

Retrieved Cases

Query

1

3

Repository Management

2

4

5

Query Processing

4 Repository

Process Variant Repository

Figure 1. Reference Architecture of PVR in [5].

In addition, we have implemented the instance adaptation framework in [7] to ensure that variants creation is organized and querying process is feasible. We also adopt the PVR architecture in [5] as shown in Figure 1 as a structured storage that consists all of the previous process designs to smooth up

18

B. Process Variants Structuring The process variants potentially can be very large and harder to be managed properly. Thus, we propose a preprocessing step through which variants will be restructured. Input graph G Output reduced graph G procedure REDUCE(G) lastsize size[G] + 1 while lastsize > size[G] do lastsize size[G] /* Terminal, sequential and adjacent reduction */ for each task t T[G] do if din[t] + dout[t]  1 then delete t else if din[t] = 1 and dout[t] = 1 then toTask[top[InFlows[t]]] top[OutTasks[t]]] delete t else if din[t] = 1 and dout[t] > 1 and taskType[t] = taskType[top[InTasks[t]]] then for each flow f OutFlows[t] do fromTask[f] top[InTasks[t]] delete t else if dout[t] = 1 and din[t] >1 and taskType[t] = taskType[top[OutTasks[t]]] then for each flow f InFlows[t] do toTask[f] top[OutTasks[t]] delete t /* Closed reduction */ if lastsize < size[G] then for each task t T[G] do if dout[t] > 1 then TaskSet {} for each flow f OutFlows[t] do if taskType[t] taskType[toTask[f]] then if toTask[f] TaskSet then TaskSet TaskSet { toTask[f] } else delete f /* Overlapped reduction */ if lastsize = size[G] then for each tasks t T[G] do if taskType[t] = FS and dout[t] = 1 and din[t] > 1 then level4 top[OutTasks[t]] t top[InTask[t]] if taskType[level4] = TASK and din[level4] > 1 and taskType [ft] = TASK and dout[ft] > 1 and din[ft] = 1 then level1 top[InTasks[ft]] if type[level1] = FS and dout[level1] > 1 then Level2 InTasks[t] Level3 OutTasks[ft] if task Level2 ( taskType[task] = TASK and InTasks[task] = { level1 } and OutTasks[task] = Level3 ) then if task Level3 ( taskType[task] = FS and OutTasks[task] = { level4 } and InTasks[task] = Level2 ) then fromTask[top[OutFlows[t]]] level1 delete all task Level2 delete all task Level3

Figure 3. Example of Business Process for Vendor Performance Assessment

 Each task tT is mapped onto a place Et and a transition Ct.  Each synchronizer sS and each task object xX such that x

 

Trig s, a place with the name  x,s is created. Synchronization achieved by creating a transition Hs which has all these places as input places and has as output places the places corresponding to the task objects triggered by that synchronizer. Each fork dD is mapped to a place Ed and has for each of its split eX an arc to a unique transition Gd,e which has an outgoing arc to Ee. Finally the initial marking of the net is a marking with one token in each of the places Ei with i an initial item.

Figure 4. Business Process of Vendor Performance Assessment using Petri nets and FCPN Approach.

We also adopt the Definition 1 (Process Model) and Definition 2 (Process Variant) from [8] to define the schema for a process model and process variant. Nevertheless, we still use explicit definition and symbol for coordinator type as defined earlier since we will use Petri nets approach in our examples of process model and process variants. We also make use of the query definition in [8] to describe our userdefined process query that will be structurally compared with the process variants.

Figure 5. Process Variants Structuring Algorithm

19

To perform this, we apply the rules from graph reduction technique in [4] and selective reduce technique in [6] to reduce the variants into graphs consisting only the tasks present in the process query, but still preserve the original structure of the process variant. We have produced the algorithm as shown in Figure 5 to reduce and restructure the process variants until it become similar as the process query. Figure 6 provide an example of process query, Q and a process variant, V. Figure 7 demonstrates the process restructuring results of V, using the presented algorithm.

Figure 8. Overview of Structural Similarity Classification

We have classified four specific classes to identify the different type of structural similarity for slightly similar category. The first class known as Exact Task and Same Construct (ETSC) grouped the process variants which have the entire task similar with the tasks in process query and similar structure construct after gone through the restructuring procedure. Another class called Exact Task and Different Construct (ETDC) is almost similar with ETSC class but the structural construct of the process variants are different from the process query. Two other classes are known as Slightly Similar Task and Same Construct (STSC) and Slightly Similar Task and Different Construct (STDC) are a bit different from ETSC and ETDC as the tasks in process variants are not completely as same as the process query after restructuring.

Figure 6. Example of Process Query, Q and Process Variant, V

C. Structural Similarity Degree In this section, we propose a computation formula scheme based on selected basic structural elements (i.e. the extra and missing task, fork and synchronizer) to demonstrate a simple preliminary result for structural similarity analysis. We make use of the justification rationale of dissimilarity weight assigned to every type of structural elements [8] which is not similar with the structure elements in process query. As a starting point, we have come out with a simple but effective computation formula that can be implemented for slightly partial structural similarity as shown in Figure 9.

Figure 7. Process Variant, V Restructuring

Variants may have different levels of similarity with the given query. In previous example, process variant, V is slightly similar to process query, Q. Several structure elements (i.e. sequential tasks, a fork and a synchronizer) in V model are reduced until V become similar as Q. In our approach, we classify the type of structural similarity based on the similarity and dissimilarity of structural elements between the process variants. If all the tasks and structure construct are exactly similar as the process query, then, it is classified as exact similar but if only some tasks or structures are similar, then it is classified as slightly similar. Detail classification for slightly similarity is required to precisely identify the type and the degree of similarity between the structure aspect of process variants and a given process query. Figure 8 provides an overview of the structural similarity classification in a hierarchical order.

Formula 1: Structural Similarity Computation for Different Structure Elements 1) matchFlow: for each task t T[P], taskType[t]  {task, coordinator} do if InFlows[t] F[Q] then count count + 1 end if if OutFlows[t] F[Q] then count count + 1 end if matchFlow = 100% *(count / F[P]) 2) matchTask: if T[P]  T[Q] > 0 matchTask = (#(T[P]  T[Q]) / T[P] )* 100% end if

20

V.

3) extraTask: if taskType[t] = task extraTask = (#(t[P]-T[Q]) *0.5 / T[P] end if

RESULT

This section demonstrates the preliminary result using the overall techniques and computation formula from our conceptual approach of structural similarity analysis.

4) extraFork: if taskType[t] = coordinator and coordinatorType[t] = fork extraFork = (#(t[P]-T[Q]) *0.8 / T[P] end if

The result for exact similar should be 100% since every task and structure construct between the process variants from this class are exactly similar to process query. Meanwhile for slightly similar case, we will apply the simple computation formula introduced in previous sub section 4.3 to compute the different type of structural similarity degree between the variants. To demonstrate this, we provide a process query, Q1 as a user-defined process query. The Figure 11 shows the structure model of Q1.

5) extraSync: if taskType[t] = coordinator and coordinatorType[t] = Synchronizer extraSync = (#(t[P]-T[Q]) *1.0 / T[P] 6) missingTask: for T[Q] - T[P] do if taskType[t] = task missingTask = (#(t[Q]-T[P]) *1.5 / T[Q] end if 7) missingFork: if taskType[t] = coordinator and coordinatorType[t] = fork missingFork = (#(t[Q]-T[P]) *1.8 / T[Q] end if 8) missingSync: if taskType[t] = coordinator and coordinatorType[t] = Synchronizer missingSync = (#(t[Q]-T[P]) *2.0 / T[Q] end if

Figure 9. Structural Similarity Computation for Different Structure Elements

It will be more professional if we could rank the structural similarity degree based on the structural similarity class described earlier. The result should be in a logic manner (i.e. the highest rank should be the exact similar, followed by ETSC, ETDC, STSC and STDC). The ranking formula for each structural similarity classification is shown in Figure 10.

Figure 11. Example of Process Query 1, Q1

We also present five process variants as shown in Figure 12 (i.e. V1, V2 and V3) and in Figure 13 (i.e. V4 and V5). These variants will be structurally compared with Q1 using our effective approach. Please note that all examples are focus on slightly similarity computation only because the rank computation for exact similar is always 100%.

Formula 2: Ranking Computation Structural Similarity For Total Match = matchFlow For Partial Match: ETSC = (matchFlow + matchTask) – ((extraTask + extraFork + extraSync) * (# (T[P]-T[Q])/ (T[P] *100%)) ETDC =

(matchFlow + matchTask) – ((extraTask + extraFork + extraSync) * (# (T[P]-T[Q])/ (T[P] *100%)) – ((missingFork + missingSync) * (# (T[Q]-T[P])/ (T[Q] *100%))

STSC and STDC = (matchFlow + matchTask) – ((extraTask + extraFork + extraSync) * (# (T[P]-T[Q])/ (T[P] *100%)) – ((missingTask + missingFork + missingSync) * (# (T[Q]-T[P])/ (T[Q] *100%)) Figure 12. Figure 10. Example of Process Variants, V1, V2 and V3

Figure 10. Rank Computation of Structural Similarity

21

between process variants in a well-structured manner. We have proposed an improvement for process reduction algorithm and Computation formula to produce a reasonable ranking percentage for the structural similarity degree of process variants. The notions and approach presented in this paper in particular has potential for further assisting the similarity analysis in the future as it could be an alternative reference method. Moreover, the results from the proposed method can improve the process design and redesign in the future ensuring organization wide consistency. Furthermore, we intend to refine and enhance the presented approach and computation formula with the intention to make it more reliable and intuitive to be applied in a larger framework with diverse dimension of process variants. REFERENCES [1]

W. M. P. van der Aalst et al. (2003) Workflow mining: A survey of issues and approaches. Data & Knowledge Engineering 47: 237-267. [2] W. M. P. van der Aalst, A.K. Alves de Madeiros and A. J. M. M. Weijters (2006) Process Equivalence: Comparing Two Process Models Based on Observed Behaviour. BPM 2006. vol. 4102, pp. 129-144. [3] W. M. P. van der Aalst (1998) The Application of Petri nets to Workflow Management. The Journal of Circuits, Systems and Computers 8: 21-46. [4] W. Sadiq and M.E. Orlowska (2000) Analyzing Process Models using Graph Reduction Techniques. Information System. Vol. 25, Issue 2, pp.117-134. [5] R. Lu, S. Sadiq (2007) A Reference Architecture for Managing Business Process Variants. Proceedings of the 9th International Conference on Enterprise Information Systems (ICEIS2007). Funchal, Portugal, 2007. [6] R. Lu, S. Sadiq (2006) Managing Process Variants as an Information Resource. 4th International Conference on Business Process Management (BPM2006). Vienna, Austria, 2006. [7] S. Sadiq, W. Sadiq, M.E. Orlowska (2005) A Framework for Constraint Specification and Validation in Flexible Workflows. Information Systems. Vol. 30, Issue 5, July 2005. [8] N. M. Mahmod, R. Lu, S. Sadiq (2008) Similarity Matching of Business Process Variants. Proceedings of the 10th International Conference on Enterprise Information Systems (ICEIS 2008). Barcelona, Spain, 2008. [9] W. M. P. van der Aalst and A. H. M. ter Hofstede (2000) Verification of Workflow Task Structures: A Petri-Net-Based Approach. Information Systems 25: 43-69. [10] F. Baccelli, S. Foss, and B. Gaujal (1995) Free-Choice Petri Nets: the Algebraic Approach. In: Proceedings of the 34th Conference on Decision & Control. New Orleans, 1995, pp. 2023-2028. [11] N. M. Mahmod (2007) Similarity Matching of Business Process Variants. Master Engineering Thesis, The University of Queensland. [12] S. Mahadevan (2004) Implementation of a consistency Test for FreeChoice Signal Transition Graphs. Master Thesis, Institute for Formal Methods in Computer Science, Software Reliability and Security Group, University of Stuttgart.

Figure 13. Extra Example of Process Variants, V4 and V5

Table 1 presents a set of preliminary result of structural similarity degree computation and ranking between V1, V2, …, V5 against Q1 using our approach for structural similarity analysis and computation formula. TABLE I RESULT OF STRUCTURAL SIMILARITY DEGREE COMPUTATION Process Variant V1

Classification of Structural Similarity ETSC

Rank 83.59%

V2

ETDC

44.28%

V4 V3 V5

STSC STDC STDC

28.86% 22.04% 13.47%

Based on the rank result in Table 1, process variant V1 carries the highest structural similarity rank which is 83.59% followed by V2, V4, V3 and lastly V5 which carries 13.47%. The reason that drive the above ranking result can be observed from the reduction algorithm and computation formula demonstrated in the previous section 4. Through observation, the overall result is sensible according to their type of structural similarity classification. VI.

CONCLUSION

Throughout this paper, we have presented an effective approach to analyse and measure the structural similarity

22


Identification of Magnetizing Inrush Current in Power Transformers using GSA Trained ANN for Educational Purposes M. Taghipour1, A. R. Moradi1, and M. Yazdani-Asrami2 1

Department of Electrical Engineering University of Birjand Birjand, IRAN

2

Department of Electrical and Computer Engineering Babol University of Technology Babol, Mazandaran Province, IRAN introduced in [3]. In this paper, the magnetizing inrush model has been derived from the structural parameters of the transformer. Also, Experiments and simulations have been carried out to verify the accuracy of proposed technique. The simulation of artificial neural network to identify magnetizing inrush current patterns in transformers has been investigated in [4]. The Hetero-associative back-propagation model has been chosen for this study. The inrush current harmonics are captured using a FFT analyzer from a transformer and used for training of the network. Then, the network was simulated in C language program. Also, Preliminary tests showed the simulated network can detect and discriminate most of the inrush current patterns after some supervised training. Also, neural network based schemes for protection of a single-phase power transformer have been investigated in [5], while applications of the neural network for a three-phase power transformer protection have been reported in [6-8]. In engineering education supporting theoretical background with practical studies provides long-term learning of the knowledge [9]. Also, it is well known that retention rate effectiveness depends on the learning experiences and the media that was used during instruction, that is, to say, the learning methodology. Retention rate is a measure of the effectiveness in promoting student retention of the material taught. The retention rate for students who practice by doing is higher than in other learning systems (lecture, reading, demonstration, or discussion group) [10, 11]. In this paper, a novel method has been used for identifying inrush current in power transformers for educational purposes. In sake of that, Multi-Layered Feed-Forward Neural Network has been utilized. In order to train ANN, a swarm-based algorithm which is called Gravitational Search Algorithm (GSA) has been run beside ANN. Also, it is mentionable that input data of this ANN are discrete samples of inrush and normal current of transformers.

Abstract- Inrush current in transformers is generated when transformer cores are driven into saturation during no-load energization. This current consists of high amplitude, large DC component and also, has much 2nd harmonic content. In the proposed paper, a new computer-aided simulation technique for teaching inrush current principles and its discrimination from normal current based on artificial intelligence has been introduced. This method can be used for educating concepts of inrush current and its identification techniques during undergraduate curriculum as an excellent approach. Evaluation of the proposed approach with undergraduate senior students is very useful in terms of their understanding of the inrush current concepts. Keywords: Artificial Neural Network (ANN), Gravitational Search Algorithm (GSA), Magnetizing Inrush Current, Transformer.

I. INTRODUCTION Power transformers are used as fundamental equipment in power system, widely. Since the failure of power transformer lead to disturbance of the power network, correct and reliable operation of protection systems has important role in continues utilization of power transformers in electrical networks. But, some of these relays encounter malfunction and inadvertent operation, because of certain transient events such as inrush currents. In general, Transformer inrush currents are high-magnitude, harmonic-rich currents generated when transformer cores are driven into saturation during energization. Therefore, inrush current recognition leads to proper operation of protection equipments. For instance, a methodology for calculation of magnetizing inrush current in frequency domain has been used in [1], because simulation of such cases in time domain has many problems. The solution is obtained by resorting to operational matrices and it is completely as the same as which are produced by time domain simulations. Also, this paper is established on a numerical method for integrating differential equations based on the concept of orthogonal approximation of functions. An analytical formula to calculate the peak inrush current of a nonlinear inductor with a series resistor has been introduced in [2]. Then, simulation results verified by comparison with measured data. A circuit model for the transient period of inrush current in a single phase transformer has been

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

II. PRINCIPLE OF INRUSH CURRENT In an electric circuit, the magnetizing transient inrush current occurs when a transformer is switched on. The peak value of inrush current depends on various factors including these parameters: the B-H characteristics of the iron core, the peak of voltage and its phase angle at the instant of switching,

23

gravitational constant that is a function of two controlling parameters. The form of G (t) is as follow:

the resistance of the primary winding, and the magnitude, and more importantly, the polarity of the residual magnetic flux density in the core at the instant of switching. An uncontrolled inrush current may lead to the inadvertent operation of the circuit’s over-current protection systems. Furthermore, the magnetic stress produced by the inrush current may damage transformer's mechanical structure and windings. In addition, this current has undesirable effects on electricity quality, extra loss and transformer life, hence, the inrush current must be reduced to overcome these defects [12, 13]. The phenomenon of transient inrush current can be explained as follows. Ignoring the winding’s resistance value, the relationship between the voltage E m Sin (ωt ) and flux φ (t ) is

-at

G(t) = G0 e T

(5)

Where, t is the current iteration of algorithm and T is the total maximum iteration of GSA .The G0 and α are known as GSA controlling parameters and they are constants. Total force that acts on ith agent in a dth dimension is calculated as follow: N

∑

Fi d (t ) =

given by (1):

rand j Fi j d (t )

(6)

j =1 j ≠i

d φ (t ) dt

(1)

Em Cos (ω t ) + φ r ωN 1

(2)

E m S in (ω t ) = N 1

φ (t ) = −

For giving a stochastic characteristic to the algorithm, the total force in dth dimension have been considered as random weighted sum of dth components of the forces, where randj is a random number in the interval [0, 1]. Each mass has a velocity and an acceleration which will be expressed as Vi(t) and ai(t), respectively. The current velocity of any mass is equal to the sum of the fraction of its previous velocity and the variation in the velocity. Variation in the velocity or acceleration of any mass is equal to the force acted on the system divided by mass of inertia.

Where, E m = φm and also, N 1 and φr are the number of ωN 1 turns in the primary windings and the remnant flux, respectively [13], [14]. During the period of transient inrush current, the transformer’s core enters into a state of saturation normally. In this coresaturated state, the magnitude of permeability would be regarded as the absolute permeability, and then the magnitude of inductance is reduced.

ai d (t ) =

Fi d (t ) M i (t )

v i d (t + 1) = rand ×v i d (t ) + ai d (t )

(7)

(8)

III. CONCEPTS OF GRAVITAIONAL SEARCH ALGORITHM When acceleration and velocity of each mass are calculated, the new position of the masses could be considered as follow:

The Gravitational Search Algorithm is a swarm-based and also is memory-less optimization algorithm based on the law of gravity. In GSA, agents are considered as objects and their performance which will be calculated by using a fitness function are expressed by their masses. The position of the each object corresponds to a solution of the problem [15]. In a system with N agents (masses), the positions are defined as follow:

x i d (t + 1) = x i d (t ) + v i d (t + 1)

New positions mean new masses. The gravitational and inertial masses are updated by the following equations:

m i (t ) = 1

d

n

X i = (x i ,...., x i ,...., x i ) For i=1, 2,…, N

(9)

(3)

fit i (t ) − worst (t ) best (t ) −W orst (t )

M i (t ) =

At a specific time /iteration (t), the force acting on ith mass from jth mass is defined as follow:

m i (t )

(10)

(11)

N

∑m

j

(t )

j =1

Fij d (t ) = G (t )

M pi (t ) × M aj (t ) R ij (t ) + ε

(x j d (t ) − x i d (t ))

(4)

Where fiti(t) represents the fitness value of the ith agent at iteration t, and, worst (t) and best (t) are defined as follow: For a minimization problem:

Where Mi and Mj are the masses related to ith agent and jth agent respectively , G(t) is gravitational constant at time /iteration (t) , ε is a small constant, and Rij(t) is the Euclidian distance between agents ith and jth agents. And G (t) is

Best (t) = min {fiti(t)} Worst (t) = max {fiti(t)}

24

IV. TRANING ANN WITH GSA

location of masses in solution space is determined randomly, that each mass has w dimensions.

In this approach, in order to train ANN the new kind of heuristic algorithm has been used which is based on swarm intelligence. In this paper, Mean Squared Error (MSE) performance function that is a criterion of difference between actual output of ANN and target output has been used as a function which should be minimized. Hence, value of MSE is reduced until zero using GSA. When MSE becomes zero, it means that actual output is same as target output and we can understand ANN has been well trained. In order to applying GSA to train ANN some steps should be done as it can be seen in Figure (1) and each step has been explain in bottom:

Step 2) applying optimizing algorithm GSA In this step, GSA algorithm has been run to train MLFFNN. In order to use GSA, value of MSE should be loaded; therefore, command feval has been used. In this part which is related to MLFFNN, this structure is created. It consists of some steps as it can be seen in follow. At first input data, target matrix and matrix of weights that expresses initial position of masses are loaded. Dimension of weight matrix is 1×w as follow: W 1 =  P1 P2 ... Pw 

(13)

Where, W1 is position of first mass in solution space which expresses weights of MLFFNN of first mass. After dividing this matrix into sub-matrixes which express weights of one layer and its biases, structure of MLFFNN is made like bottom: S 1 = W θ × [ID ] + b1 (14)

S 2 = log sig (S 1 )

(15)

Z 1 = W β × [S 2 ] + b 2

(16)

Z 2 = log sig ( Z 1 )

(17)

O1 = W γ × [ Z 2 ] + b3

(18)

O 2 = log sig (O1 )

(19)

Fig 1. Flowchart of training ANN with GSA

Where, ID is input data, W θ ,W β and W γ are weights between input and first layer, between first layer and second one and between second layer and third one, respectively. b1, b2 and b3 are biases related to first layer, second layer and third layer, respectively. Also, training processes of GSA require a bounded and differentiable activation functions. Therefore, sigmoid function has been used. At the end, O2 is actual output of network and MSE value of subtract of actual output and target output is calculated , this value is fitness value of MLFFNN for GSA and proposed algorithm will change weights till this value becomes minimum in next iterations. MSE function can be seen in equation (20).

Step 1) Determining initial parameters: In this step, number of layers and number of neurons in each layer and kind of ANN are determined. After that initial parameters of GSA are determined which are the number of masses/agents, the number of problem/solution dimension which is dependent on number of synaptic weights of MLFFNN, the number of weights or in other word number of solution dimension which can be calculated by equation (12). (Number of Input Data ) × θ + b1 + θβ + b 2 + βγ + b3 = (12) ω (Number of Neural Network weights )

Where, θ, β and γ are the number of neurons in first layer, second layer and third one, respectively. And b is bias for each neuron. Then value of controlling parameters G0 and α and number of iteration are determined, eventually in this part primary

∑ (target output - acual output) MSE =

25

2

(20)

M ,N

M ×N

Step 3) updating GSA parameters in this step, parameters of GSA are updated, at first acceleration is updated then velocity and position of masses will be updated, as mentioned before, these new position for one mass expresses new weights for MLFFNN. Step 4) inspection of stop criterion In order to stop this algorithm, there exist two criterions; one is number of iteration and other is MSE value, whenever this value becomes zero, it means that actual output and target output are same.

Fig 2. Block diagram of simulation of identifying Inrush current

Step 5) End V.

UTILIZATION OF PROPOSED APPROACH TO IDENTIFY INRUSH CURRENT

As we know, it is impossible to apply analog signal to ANN, hence, discrete sample data have been extracted from inrush and normal current waveforms, and in order to have valid data, and data in [16] has been used, but because of severe variation in data they cannot be used without any processing, therefore, this data has been normalized as follow:

data − min(row ) normalized data = max(row ) − min(row )

Fig 3. Curve of MSE value Number of mass= 10, α = 15 and G 0 = 200 TABLE I Test results for train data and test data

(21)

Test data 5MVA, 66/33kV 3MVA, 66/33kV

Normalized data of simulated transformers have been applied to MLFFNN as input data, also two classes have been considered as output for normal condition and inrush current. In output "01" is normal condition and "10" is inrush current situation. In Figure (2), block diagram of simulation can be seen.

2MVA, 66/33kV

Actual output

Target output

0.0000 1.0000 0.0000 1.0000 0.0000 1.0000

0.0000 1.0000 0.0000 1.0000 0.0000 1.0000

1.0000 0.0000 1.0000 0.0000 1.0000 0.0000

1.0000 0.0000 1.0000 0.0000 1.0000 0.0000

Error 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000 0.0000

Table II Result analysis for different magnitude of α

Magnitude of α

VI. RESULTS AND DISCUSSION

α = 10 α =20

In order to reach our education purpose, data of 12 transformers out of 15 have been chosen as train data and other were test data. Also, for this work controlling parameters are; α = 15 and G 0 = 200 , and number of masses is equal to "10", in other hand, MLFFNN has one input layer with 16 neurons, two hidden layers with 16 and 32 neurons and one output layer with two neurons based on two output classes. At first, ANN has been trained with train data then test data have been used to test trained ANN, in Table (I), result data or in other word actual output of MLFFNN for test data has been listed. At result, by using proposed algorithm, MSE value becomes 5.1191e-010 and its values' curve has been shown in Figure (3), this figure shows that algorithm is stopped at iteration 500, that is, algorithm arrive at one of the stop criterions, also after 30 times run, it has been observed that mean value of MSE is about 9.9098e-010 .from results, it can be understood that proposed algorithm can deal with identifying inrush current very well and can do this work with negligible error.

α = 25

Target output 1.0000 0.0000 1.0000 0.0000 1.0000 0.0000

0.0000 1.0000 0.0000 1.0000 0.0000 1.0000

VII.

Actual output 1.0000 0.0000 1.0000 0.0003 1.0000 0.0031

0.0000 1.0000 0.0000 0.9999 0.0004 0.9949

Error 0.0000 0.0000 0.0000 0.0003 0.0000 0.0031

0.0000 0.0000 0.0000 0.0001 0.0004 0.0061

MSE of trained MLFFNN 9.1039e-011 6.4785e-008 8.1769e-005

SENSIVITY ANALYSIS

In order to educational purpose, sensitivity of results to variation of GSA parameters such as controlling parameters α and G0 has been analyzed. This work has been done for 2MVA, 66/33kV transformer, that is, for each condition; trained MLFFNN in different conditions has been used for testing this transformer. Therefore, in Table (II), (III) sensitivity to α and G0 have been shown, respectively, also time of testing mentioned transformer has been calculated, and after obtained following results, it is became clear that test time for one transformer by proposed approach is about 15 ms and it seems good time for identifying Inrush current. Also reported MSE value is best in 30 times run and it should be mentioned that in this 30 times these values of MSE had approximately same value.

26

Table III. Magnitude of G0 G0= 100 G0=150 G0= 250

Target output 1.0000 0.0000 1.0000 0.0000 1.0000 0.0000

[6] M. R. Zaman and M. A. Rahman, "Experimental testing of the artificial neural network based protection of power transformers," IEEE Transactions on Power Delivery, vol. 13, no. 2, April 1998, pp. 510– 517.

Result analysis for different Magnitude of G0

0.0000 1.0000 0.0000 1.0000 0.0000 1.0000

Actual output 1.0000 0.0052 1.0000 0.0027 1.0000 0.0000

0.0001 0.0000 0.0011 1.0000 0.0000 0.9999

Error 0.0000 0.0052 0.0000 0.0027 0.0000 0.0000

0.0001 0.0000 0.0011 0.0000 0.0000 0.0001

MSE of trained MLFFNN 8.9099e-009

[7] P. L. Mao, and R. K. Aggarwal, "A Novel Approach to the Classification of the Transient Phenomena in Power Transformers Using Combined Wavelet Transform and Neural Network," IEEE Transactions on Power Delivery, Vol. 16, No. 4, October 2001, pp. 654-660.

1.1807e-008 5.5637e-010

[8] A. Chatterjee, M. Maitra, and S. K. Goswami, "Classification of Over-Current and Inrush Current for Power System Reliability using Slantlet Transform and Artificial Neural Network," Expert Systems with Applications, Vol. 36, 2009, pp. 2391–2399.

Similar to previous analysis, finding optimum α is so important that can help designer to design best program to solve problem, as it is observed by increasing magnitude of α accuracy of approach and MSE value are reduced especially in identifying Inrush current and the range which is proposed is 7 to 20.In this analysis G 0 = 200, number of masses = 10 and

[9] E. Tanyildizi and A. Orhan, A Virtual Electric Machine Laboratory for Effect of Saturation of the Asynchronous Machine Application, Comput Appl Eng Educ, Vol. 17, No. , 2009, pp. 422-428.

structure of MLFFNN is 16-16-32-2. By testing for different magnitude of G 0 optimum value of

[10] P. Ramsden, Learning to teach in higher education, 2nd edition, Routledge, Abingdon, 2003.

this parameter for this problem is in interval [180-280]. In this analysis number of masses = 10, α = 15 and structure of MLFFNN is 16-16-32-2.

[11] C. Furse, Teaching and learning combined (TLC), IEEE Antennas Propag Mag, Vol. 45, 2003, pp. 166-167. [12] M. Reza Feyzi and Dr. M. B. B. Sharifian, "Investigation on the Factors Affecting Inrush Current of Transformers Based on Finite Element Modeling," IEEE (IPEMC 2006), 2006.

VIII. CONCLUSION

[13] M. Yazdani-Asrami, A. Ebadi, R. Ahmadi Kordkheili, M. Taghipour, "Effect of Null Wire on the Peak Value of Inrush Current in Three-Phase Transformers Bank," International Review on Modelling and Simulations (IREMOS), Vol. 3, No. 2, April 2010, pp. 140-145.

Inrush current can be harmful for network, therefore, proper and quick identifying this unexpected current is so important, by reviewing results, it can be understood that ANN can be a good tool for classifying these kinds of malfunctions and proposed algorithm is very suitable and proper for training ANN and it can help engineers to design excellent classifier. According to results, choose best controlling parameters for GSA and also optimum number of masses, besides, using optimum number of layers and neurons are so important for reaching best results.

[14] Shin-Der Chen, Ray-Lee Lin, and Chih-Kun Cheng, “Magnetizing inrush model of transformers based on structure parameters” IEEE Transactions on Power Delivery, Vol.20, No.3, 2005, pp. 1947–1954. [15] E. Rashedi, H. Nezamabadi-pour, and S. Saryazdi, "GSA: A Gravitational Search Algorithm," Information Sciences, Vol. 179, 2009, pp. 2232–2248. [16] M. Geethanjali, S. Mary Raja Slochanal, and R. Bhavani, "PSO trained ANN-based differential protection scheme for power transformers," Neurocomputing, Vol. 71, 2008, pp. 904–918.

ACKNOWLEDGMENT The authors would like to thank Mr. Ali Darzi (University of Birjand) for his important help. REFERENCES [1] J. J. Rico, E. Acha, and M. Madrigal, "The Study of Inrush Current Phenomenon Using Operational Matrices," IEEE Transactions on Power Delivery, Vol. 16, No. 2, April 2001, pp. 231-237. [2] Y. Wang, S. G. Abdulsalam, and W. Xu, "Analytical Formula to Estimate the Maximum Inrush Current," IEEE Transactions on Power Delivery, Vol. 23, No. 2, April 2008, pp.1266 – 1268. [3] S. D. Chen, R. L. Lin, and C. K. Cheng, "Magnetizing inrush model of transformers based on structure parameters," IEEE Transactions on Power Delivery, Vol.20, No.3, 2005, pp. 1947–1954. [4] C. T. Wai, C. C. Keong, and G. H. Beng, "Detection of Magnetizing Inrush Current Using Artificial Neural Netork," IEEE TENCON 1993 / Beijng [5] G. Perez, A. J. Flechsig, J. L. Meador, and Z. Obradovic, "Training an artificial neural network to discriminate between magnetizing inrush and internal fault," IEEE Transactions on Power Delivery, vol. 9, no. 1, January 1994, pp. 434–441.

27


Making DC-DC Converters Easy to Understand for Undergraduate Students Mohammad Yazdani-Asrami Department of Electrical and Computer Engineering Babol University of Technology Babol, Iran [email protected]

Reza Ahmadi Kordkheili Department of Electrical and Computer Engineering Babol University of Technology Babol, Iran [email protected]

Amir Mohammad Sayidi Department of Electrical and Computer Engineering Science and Research Branch of Islamic Azad University Tehran, Iran [email protected] based on a voltage or current divider, are inefficient. This is because they are limited to output voltages smaller than the input voltage, and also, considering their requirement to low-frequency (50 or 60 Hz) line transformers and filters, their power density is low. Linear regulators can, of course, provide a very high-quality output voltage. Their main area of application is at low power levels. Electronic devices in linear regulators operate in their active (linear) modes, but at higher power levels switching regulators are used. Switching regulators use power electronic semiconductor switches in on and off states. Because there is a small power loss in those states (low voltage across a switch in the on state, zero current through a switch in the off state), switching regulators can achieve high energy conversion efficiencies. Modern power electronic switches can operate at high frequencies. The higher the operating frequency, the smaller and lighter the transformers, filter inductors, and capacitors. In addition, the dynamic characteristics of converters improve with increasing operating frequencies. The bandwidth of a control loop is usually determined by the corner frequency of the output filter. Therefore, high operating frequencies allow for achieving a faster dynamic response to rapid changes in the load current and/or the input voltage [4-6]. High-frequency electronic power processors are used in dc-dc power conversion. The functions of dc-dc converters are [4]: 1) To convert a dc input voltage VS into a dc output voltage VO; 2) To regulate the dc output voltage against load and line variations; 3) To reduce the ac voltage ripple on the dc output voltage below the required level; 4) To provide isolation between the input source and the load (isolation is not always required); 5) To protect the supplied system and the input source from electromagnetic interference (EMI); 6) To satisfy various international and national safety standards; The input to these converters is often an unregulated dc voltage, which is obtained by rectifying the line voltage, and therefore it will fluctuate due to changes in the linevoltage magnitude.

Abstract—Power electronic laboratory may seem to be a good method of education. Despite the benefits of a laboratory-based education, some serious limitations make it difficult to consider this method as a main educating tool. In addition, power electronic laboratory is one of the most expensive laboratories in both foundation and operation. Moreover, the complexity of power electronic circuits makes it hard to repeat the test for different conditions. The increasing number of students accessing university educational program make the problem even more complicated. Adding probable risks of electrical accidents in laboratory for students, breaking down of circuit equipment due to misuse, limited access to the laboratory for students can clarify other aspects of problem. To overcome such problems, computer-based education have been widely developed and used in universities. Different software programs with different features have been developed to help understanding the concepts and operation of different power electronic circuits. One of the most effective software programs in education context is MATLAB/SIMULINK software, an easy to used software with so many capabilities and tools. Thus, to make students more familiar with power electronic circuits and show the effect of different parameters on the behavior and output of these circuits, our educational classes were based on MATLAB/SIMULINK software. This paper covers "buck-boost" and "cuk" converters taught and simulated in power electronic course. Keywords — DC-DC Converters; MATLAB/SIMULINK; Power Electronic Education

I. INTRODUCTION Appearance of power electronic devices has had an undeniable effect on electrical and control systems. Considering various capabilities of these devices, many researchers and industries are interested in utilizing these devices to improve the efficiency of systems. Nowadays, wide spread use of power electronic devices and circuits makes the education a necessity for undergraduate students. Education is also a demand of industries, where power electronics has found many different applications [1-3]. Modern electronic systems require high-quality, small, lightweight, reliable, and efficient power supplies. Linear power regulators, whose principle of operation is

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

28

Simulation based educations have become one of the main methods to help teaching and speeding up the learning process in many education fields such as signal processing, power electronics, electromagnetic fields, frequency response analysis, and power systems, among others [7-11]. With a suitable simulator analyze, design, and development of power electronics circuits could be more convenient, shorter and low-cost. In a computer simulation based education of power electronics, all currents and voltages of elements can be traced. In addition, on the contrary laboratory environment, with the help of computer simulation, the effect of the change in circuit parameters on the performance of the circuit can be examined easily. On the other hand, the student can manipulate the circuit configuration during the application and see the results immediately. Since the applications are based on the simulation the mistakes and faults will not give any harm. Because of all these reasons, using suitable simulation software during the education process may help students to learn, develop and improve the education quality of power electronic course. Several commercial programs can be used for the simulation of power electronic circuits such as ELECTRONICS WORKBENCH, PSPICE, LABVIEW, PSIM and EMTP. Also, some educational tools have been used in recent years such as POWERLAB in [12]. However, due to some disadvantages such as complicated structure for learning, commercially difficulty, high oscillations and etcetera, they may not be significantly useful in the education process of power electronic course. So, in this paper, a MATLAB-SIMULINK based technology for costeffective education, training and simulation of power electronic dc-dc converters is presented. The following converters are discussed in this paper: 1) Buck-Boost Converter 2) Cuk Converter II.

B. Cuk Converter The circuit of the cuk converter is shown in Fig.2. It consists of dc input voltage source VS, input inductor L1, controllable switch S, energy transfer capacitor C1, diode D, filter inductor L2, filter capacitor C, and load resistance R. An important advantage of this topology is a continuous current at both the input and the output of the converter. Disadvantages of the cuk converter are a high number of reactive components and high current stresses on the switch, the diode, and the capacitor C1. When the switch is on, the diode is off and the capacitor C1 is discharged by the inductor L2 current. With the switch in the off state, the diode conducts currents of the inductors L1 and L2, whereas capacitor C1 is charged by the inductor L1 current. The dc voltage transfer function of the cuk converter is: VO D (2) =− VS 1− D This voltage transfer function is the same as that for the buck-boost converter.

Figure 1. The block diagram of buck-boost converter

DC-DC CONVERTERS AND THEIR CONTROL

A. Buck-Boost Converter A simple topology of buck-boost converter is shown in Fig.1. The converter consists of dc input voltage source VS, controlled switch S, inductor L, diode D, filter capacitor C, and load resistance R. With the switch on, the inductor current increases while the diode is maintained off. When the switch is turned off, the diode provides a path for the inductor current. Note the polarity of the diode that results in its current being drawn from the output. The buck-boost converter waveforms are presented in simulation section. The condition of a zero volt-second product for the inductor in steady state yields: VO D =− VS 1− D

Figure 2. The block diagram of cuk converter

C. Control of DC-DC Converters In a dc-dc converter with a given input voltage, the average output voltage is controlled by controlling the switch on and off durations (ton and toff). One of the methods for controlling the output voltage employs switching at a constant frequency and adjusting the on duration of the switch to control the average output voltage. In this method, called pulse-width modulation (PWM) switching, the switch duty ratio D is varied [5]. The switch is being operated with a duty ratio D defined as a ratio of the switch on time to the sum of the on and off times [4]. For a constant frequency operation:

(1)

The output voltage VO is negative with respect to the ground. Its magnitude can be either greater or smaller (equal at D= 0.5) than the input voltage as the name of the converter implies. The structure of the output part of the converter is similar to that of the boost converter (reversed polarities are the only difference) [4-6].

D=

29

t on t on + t off

(3)

III. SIMULATION RESULTS III. As mentioned, MATLAB/SIMULINK software is used to simulate two mentioned converters. This part presents SIMULINK circuits and the outputs obtained by software. Furthermore, to verify the obtained outputs, all figures are compared with the figures given in reference [5], as a world world--wide wide reference. It should be noted that the input dc voltage (Vd) is assumed to be 100volt to ease tthe he comparison. A. A. Simulation Switching Pulses As stated, PWM switching method is used in this program. Considering that the normal switching frequency range of dc dc--dc dc converters is about a few K KHz Hz to a few hundred KHz [4, 5], the switching frequency used in these simulations is 10 KHz. The SIMULINK block diagram of method is presented in Fig. Fig.3 3.. To make students more familiar and more convenient with software, a simple method is used for producing the sawtooth voltage. The block compares the reference signal with the produced sawtooth signal, and the result acts as switching pulses. A sample of switching pulses is presented in Fig. Fig.4 4..

Figure 4. switching pulses of IGBT

t

C ontinuous signal

IGBT switch g

m

C

E

Vout

i -

switching block

+ v -

+ v -

load

load voltage

+

inductor current

gain (-1)

i -

input voltage (Vd)

-1

+

diode

B. B. Simulation of the Buck Buck--Boost Boost Converter As the name states, the co converter nverter can act in two different modes, buck mode and boost mode. The SIMULINK block diagram is presented in Fig. Fig.55.. The The parameters values are: R=10Ω, R=10 R=10Ω Ω,, L=330e L=330e--4, 4, C=220e C=220e-6. 6. As discussed, the polarity of output voltage reverses in buck buck--boost boost converters. To make the figure of outp output ut voltage more common and easy to understand, the output voltag voltagee is multiplied by gain ((--1). 1). The SIMULINK results under each mode are presented in this part. The output voltage of converter in buck mode, presented in Fig. Fig.6 6,, verifies the ability of converter to decrease the input voltage,Vd. MATLAB Power GUI toolbox empowers us to have the FFT analysis for a specific waveform. The FFT of output voltage illustrates the output magnitude for each frequency order. This toolbox also calculates the THD of waveform, a criteri criterion on to evaluate whether the waveform is suitable en enough ough or not. The FFT of output voltage in either buck or boost mode is presented in Fig. Fig.7 7. The output voltage of converter in boost mode, together with its FFT analysis, and the inductor current are presented in Fig. Fig.99 through hrough Fig. Fig.11 11 11,, respectively. The THD value of output voltage is also presented on the top of each FFT figure. As can be seen, the THD of output voltage is 17.64% and 17.91% for buck and boost mode mode,, respectively. Comparing the two output voltage figures demonstrates the ability of converter tto o either increase or decrease the input voltage, Vd.

iL

inductor voltage

VL

Figure 5. Simulink block diagram for buck buck-boost boost converter

60

Vout (volt)

40

20

0

-20

0

0.01

0.02

0.03

0.04

0.05

Time (s)

Mag (% of DC component)

Figure 6. output voltage of buckbuck-boost boost converter in buck mode

FFT analysis of Vout (DC component = 36.56 , THD= 17.64%) 25 20 15 10 5 0

0

200

400

600

800

Frequency (Hz) Figure 3.. SIMULNK block diagram of switching switching method

Figure 7. FFT of output voltage in buck mode

30

1000

C. Simulation of Cuk Converter The block diagram of simulated circuit is shown in Fig.12. The gain block (-1) in the output of converter has the same role as in the buck-boost converter. The output voltage of converter, the current of inductor L1 and L2 in both buck and boost mode are shown in Fig.13 through Fig.18. This paper represents part of our educational program for power engineering students. Each duration consists of 4 classes, each containing 10 students. To evaluate the efficiency of method, we assessed students through some questionnaires. Results reveal students’ satisfactory with our program. Actually, a significant percent of students stated that the program has fascinated them to power electronics and its concepts and circuits, which really delighted us. There were also some objections for which we have discussed and we are trying to omit them and improve our program’s quality.

Figure 8. inductor current of buck-boost converter in buck mode

Figure 9.

output voltage of buck-boost converter in boost mode

Figure 12. Simulink block diagram for cuk converter Figure 10. FFT of output voltage in boost mode

Figure 11. inductor current of buck-boost converter in boost mode

31

350

50

300

40

Vout (volt)

Vout (volt)

250 200 150 100

30 20 10 0

50 0

0

0.01

0.02

0.03

0.04

0.05

-10

0.06

0

0.01

0.02

Time (s)

0.03

0.04

0.05

0.06

Figure 13. Output voltage of cuk converter in boost mode

Figure 16. Output voltage of cuk converter in buck mode

FFT analysis of Vout (DC component = 281.8 , THD= 12.34%)

FFT analysis of Vout (DC component = 38.58 , THD= 11.92%)



Time (s)

10 8 6 4 2 0

0

200

400

600

800

1000

Frequency (Hz)

12 10 8 6 4 2 0

1.6

70

1.4

60

1.2

iL1 (A)

1.8

80

iL1 (A)

2

90

50 40

20

0.4

10

0.2 0

0.03

0.04

800

1000

0.8 0.6

0.02

600

1

30

0.01

400

Figure 17. FFT of output voltage of cuk converter in buck mode

100

0

200

Frequency (Hz)

Figure 14. FFT of output voltage of cuk converter in boost mode

0

0

0.05

0.06

0

0.01

0.02

0.03

0.04

0.05

0.06

Time (s) Figure 18. inductor current of cuk converter in boost mode

Time (s) Figure 15. inductor current of cuk converter in boost mode

32

IV. CONCLUSION Using computer-based education in power electronics, especially softwares designed for graphically simulating such circuits is a very effective and suitable method to make power electronic issues easy to understand. MATLAB/SIMULINK software is used in our educational program to cover different concepts and areas of power electronics, such as dc-dc converters, inverters, rectifiers, cycloconverters among their applications. Due to page limitations, this paper presents only a part of educational program, "buck-boost" and "cuk" dc-dc converters. The concepts of the converters have been presented. Simulation circuits in MATLAB/SIMULINK software have been shown, and a simple switching method of these converters have been described and simulated. To understand the operation of converters, an output sample of converters has been also presented. The proposed method helps students to learn and develop power electronic circuits. Students have learnt at their own progressive rate as well as they have actively participated in more interest. It was taught during the second semester of the 2008-2009 academic year and was well accepted by the students. Evaluation of the project involving 40 students indicates benefits of this project for learning subject. By considering the students feedback for proposed education method, can be judged that MATLAB-SIMULINK is the best software for simulation and learning of power electronic course. According to authors experiences this method can act as a helpful tools in order to increase the instructors’ ability in teaching the proposed subject. The most positive result from the present simulation is such as, hardware set for laboratory test is very expensive; therefore, for those universities software method can be useful. Also, the simulation can be done in the classroom during subject presentation and can be very effective for subject understanding. In addition, by use of this simulation and education method can be preventing the possible high voltage damages of laboratory tests to students.

REFERENCES [1]

S. Tuncer, Y. Tatar, and H. Guldemir, “Design and Implementation of an Integrated Environment for Real-Time Control of Power Electronic Systems,” Comput. Appl. Eng. Educ., Vol. 17, pp. 119-130, 2009. [2] C. Elmas and Y. Sonmez, “An Educational Tool for Power Electronics Circuits,” Comput. Appl. Eng. Educ., Vol. 18, pp. 157165, 2010. [3] A. Keyhani, M. N. Marwalli, L. E. Higuera, G. Athalye, and G. Baumgartner, “An integrated virtual learning system for the development of motor drive systems,” IEEE Trans. Power Syst., Vol. 17, pp. 1-6, 2002. [4] M. H. Rashid, Power Electronics Handbook, Academic Press, San Diego, California, 2001, pp. 211-224. [5] N. Mohan, T. M. Undeland, and W. P, Robbins, Power Electronics: Converters, Applications and Design, 3rd ed., John Wiley & Sons Inc., 2003, pp. 161-199. [6] F. Mazda, Power Electronics Handbook, 3rd ed. Newnes, 2003, pp. 261-283. [7] B. Ando, S. Graziani, and N. Pitrone, Stand-alone laboratory sessions In sensors and signal processing, IEEE Trans. Educ., Vol. 47, pp. 4-9, 2004. [8] T. W. Gedra, S. An, Q. H. Arsalan, and S. Ray, Unified power engineering laboratory for electromechanical energy conversion, power electronics, and power systems, IEEE Trans. Power Syst., Vol. 19, pp. 112-119, 2004. [9] F. S. Sellschopp and M. A. Arjona, An automated system for frequency response analysis with application to an undergraduate laboratory of electrical machines, IEEE Trans. Educ., Vol. 47, pp. 57-64, 2004. [10] R. H. Chu, D. D. Chuan Lu, and S. Sathiakumar, Project-Based Lab Teaching for Power Electronics and Drives, IEEE Trans. Educ., Vol. 51, No. 1, pp. 108-113, 2008. [11] F. L. Tan and S. C. Fok, Development of a Computer-Aided Educational Tool Incorporating MATLAB for Engineering Measurements, Comput. Appl. Eng. Educ., Vol. 17, pp. 206-213, 2009. [12] O. Montero-Hernandez, A. R. De La Rosa, D. Baez-Lopez, R. Alejos, and E. Enriquez, Power Lab: A tool to Learn Electrical Machines and Power Electronics, Comput. Appl. Eng. Educ., Vol. 7, pp. 213-220, 1999.

33


Survnvote: A Free Web Based Audience Response System to Support Interactivity in the Classroom Teddy Mantoro1, Media A. Ayu2, Emir Habul, Ana U. Khasanah, INTEG Research Group, Kulliyyah of Information and Communication Technology International Islamic University Malaysia, Kuala Lumpur, Malaysia 1 [email protected], [email protected]

Abstract— Survnvote is a free Audience Response

inconvenience to use, since they need to do the set up and other technicalities [2,3]. Votapedia [7] is the first web-based ARS developed to overcome and minimize the difficulties of implementing ARS in the classroom which replaces the clicker with mobile phone and use the web instead of specialist software and infrastructure that needs to be installed. Votapedia system is an open source system, free of charge and it has been used by many users, especially the ones in the educational environment. However, the system is still not yet perfect. There are many rooms for improvement in Votapedia. One of them is the userfriendliness of the interface design for survey creation by the user. In terms of flexibility and usability, the user is unable to reuse the same survey, there is no user configuration in the survey setting to guaranty one man one vote. Survnvote has been built as an extension to Votapedia system with a better approach to provide a better ARS service to the users which use a wiki based application. We identify five contributions in this study. First, the weakness in user-friendliness interface has been looked after in this study so that user can use the panel without knowing the wiki syntax. Second, Survnvote introduces a new concept of crowd group synchronization to make a respondent only can use one way, either phone, web or sms, to cast her/his vote in the participated survey. It promotes the principle of one-man-one-vote which makes Survnvote able to provide more accurate and reliable service. Third, Survnvote give flexibility to create survey by providing more types of questions, handles mathematical equations and graphics. Fourth, the presentation of results in Survnvote give more options in displaying the data ranging from the statistics, raw data, and various types of graphs which have background that can be customized. Lastly, in order to bring the result on public presentation, the progress of the survey is able to be seen on PowerPoint slide using LiveWeb enabled plug in. Lastly, the results are also made available in .cvs and .xls extensions to make the documentation and interpretation of the data easier. This paper describes the development and the use of Survnvote. Section 2 discusses the design specifications of the Survnvote followed by Section 3 which presents an overview, including services provided and how to use

System (ARS) for surveys and voting using the web, sms and mobile devices (PDA or smart phone). This paper proposed some improved features of a web-based ARS which replace the clicker with mobile phone and use the web instead of specialist software and infrastructure. Survnvote solved the main problem of traditional ARS which is high cost and need to physically install the receivers in the room. This system offers a more user friendly approach and improved features, e.g. the introduction of crowd management, Survnvote panel that is easy to use, crowd group voting, and several ways to participate in voting. This system gives a better result representation by providing several data formats such as graph, PowerPoint, Excel. Survnvote also provides data analysis to the user. ARS users will experience valuable services by utilizing this system to support interactivity in their classrooms, seminars and conferences. A test-case of the use of this system in a lecture class is provided in this paper. Keywords — audience response system; web-based survey; open source system; classroom interactivity.

I.INTRODUCTION Electronic Audience Response System (ARS) has been around since midst nineties [1]. It has been used in the classroom setting to engage students for improving the active learning environment. Using ARS, students are able to respond to the question made by lecturer and get immediate feedback and result. The type of questions asked usually a multiple choice question, seeking correct answer for a problem, or open ended question with simple text message. Mostly, ARS uses clickers and requires specialist infrastructure and software. Some thought that this system is considered expensive to be provided, needs time to set the system, increase the stress of the audience if there is failure and it takes time to make audience familiar with the clicker [2,3,4]. Others believe that although clickers are expensive and require a good maintenance, it gives the anonymous type of response and make respondent more likely to participate and give honest response to the question posed [5,6,7,8]. However, still these benefits did not make ARS widely adopted in education environment. Some still find it is

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

34

them, which covers a brief description about the improved features provided in Survnvote system. Section 4 shows an example on how the improvement made the system performs better in providing service to the user is also provided. The paper closed by a brief conclusion in Section 5.

As for the client, it was designed by implementing several open source applications and has the following requirements (Figure 1): 1. the survey creator will log-in to Survnvote web and bring up his survey. 2. The survey will be active in front of audiences, showing the questionnaire or quiz including 3 modes of participation, i.e. through mobile phones, sms and web. 3. The participants can use their mobile phones, by sms or dial, based on their selection, or they can browse from their mobile phone and select their answer. 4. The participant can use any regular PC that is connected to the internet for selection using web based only. 5. It is also possible for the participant to use their regular hand set if they know the number that they are going to select (This is usually for long time surveys and the participants are spread out in different geological locations). 6. The survey can be closed in two ways: stopped by survey creator or by time setting. 7. During the survey, the survey creator can add more time on the fly. 8. After the survey is stopped, the creator can continue the survey. 9. The creator can re-run the same survey several times for different audiences without any need to worry that the previous survey will be missing. He/she can compare the results at the end of the survey.

II. SURVNVOTE DESIGN SPECIFICATIONS Survnvote is a further development of Votapedia which is developed by Intelligent Environment Research Group (INTEG) KICT-IIUM. The service is available on the Survnvote website (http://www.survnvote.net/). Until this stage, the system is available for web-based and SMS survey, the phone line voting is still under development, since currently the VOIP phone line in Malaysia still under our investigation to be integrated in Survnvote. Survnvote was developed using client-server approach. The server was designed to receive vote or survey data from phone lines, SMS and the internet, and at the same time to deliver survey result through the web by using several open source applications (Figure 1). The server requirements are as follows: 1. A server which connect to the internet has Survnvote application installed. 2. A mobile phone is connected to the server, to be used by Gammu smsd to receive and send sms. 3. An asterisk server is connected to 100 digital phone lines and it connects between Asterisk server, Survnvote server and campus PABX (on progress). 4. When a web server, using apache server, is running, the survnvote application is also running by utilizing several programs including Mediawiki and a plug-in Mobileskin, PHP and MySQL. 5. The survnvote server will be online and in listening mode, waiting to respond to a user request. 6. When there is a response from the participant, the server will collect the data and show immediate output in form of graphics based on current user participation. 7. When the survey has finished, the survnvote server will calculate the statistical report, preparing the csv or xls files to be exported, and record the survey.

III. SURVNVOTE: AN OVERVIEW Survnvote has several main functions to allow the user to use the 4 (four) key features of the application i.e. creating a survey, running a survey, data representation, and user registration process. The following sections describe the key features and services provided in Survnvote.

Fig.2. The front page of Survnvote A. Creating a survey There are three types of survey offered by Survnvote as shown in Figure 2. The first type is a simple survey which can create only one question with the answers.

Fig.1 : Survnvote General Architecture

35

Second is questionnaire which user can create many questions and options. And lastly the quiz type which has the same feature with the Questionnaire but the user can specify the correct answer before the survey is started and give a point (mark) for the correct respond. In all of the create survey page, there are three tabs which are New Survey, Voting Options, and Display Setting.

Phone Voting, the amount of questions and the options will be limited. It is because the phone lines that are available need to be shared with other active survey(s). When a creator opts to have the caller ID to be enabled, he/she will be able to have the information on who the respondents of the survey are. From outside Malaysia, it is recommended to use the unidentified voters since most of caller ID from other countries cannot be recognized.

New Survey tab is to create new survey. This function built to handle more survey under the same title. Survnvote gives a chance to create the same survey several times, and record the previous result neatly. The questions will be numbered automatically and the options can be adjusted, e.g. move the third option to be the first option and vice versa is possible (Figure 3). This survey creation interface is very user friendly and easy to use. The creator is not required to understand wiki syntax to edit his/her survey.

Fig.4. Voting options For a web-based voting, the respondents can cast their vote by accessing the website. If the creator wants the respondent to be identified, the setting should be changed to “enabled registered web voting” so that the website can track the unique IP address of the respondent. In order to be able to identify the respondents, they need to log in to the website before participating in the survey.

Fig.3. New Survey tab to create new questionnaire on the Survnvote website The voting option tab gives the lecturer an advance voting option (please refer to Figure 4). The survey privacy can make the survey is limited to a particular group called crowd or a public survey that everybody can participate on the survey. For the crowd member, they need to be registered and login to Survnvote website. This feature helps the lecturer to conduct the same survey for different classes and repeat it again as needed. The duration of the survey can be changed, maximum 5 hours for phone line based and 5 days for the web based survey.

Fig.5. Display Setting Survnvote display setting is very useful to produce the result of the survey and represent it in statistical and graphical display (Figure 5). The respondent can see the progress of each vote if the option is enabled. However, for the quiz type, the user can only see the results when the survey has finished. For the user satisfaction-like survey, Survnvote displays only the certain number of top result. For example, there are four brands of mobile phones and it only displays the two most popular brands. For the representation of the survey, the user can upload

Interestingly, Survnvote can limit the way respondents give the vote. The lecturer (creator of the survey) has the options to enable or disable the phone voting and web voting. This is an example of the flexibility of the Survnvote that allows users of the system to personalize their settings to their preference. The phone line voting has several limitation. Once the creator enables the

36

the additional background. The final result of the survey will be presented on the PowerPoint and the number of questions on each slide can be specified on this option tab. It gives Survnvote powerful features in producing, conducting and representing the survey.

Figure 7 shows that the survey already conducted twice, Run#1 and Run#2. The current survey is the newest survey that runs on the system. The representation of the result is not only by the colorful charts but also by giving the percentage of the final result. In “view detail” option Survnvote gives the complete representation of the data, not only for the whole result but also gives the detail of the individual question.

Survnvote system also caters surveys which need to display complex mathematical equations (Figure 5). Creator will need to use math tags to be able to display the correct mathematical formula. However, the user friendly interface provided in this system make the creation of this special display easy. Every time the user writes the script, the preview can be seen on the field which is provided below it.

Fig.8. Embed the survey to PowerPoint Another improved feature in Survnvote is enabling the creator to embed the survey that he/she has created to a PowerPoint presentation and control it from there (Figure 8). This feature makes the lecturer does not need to go back and forth from his/her PowerPoint slides to the survey web page.

Fig. 6. The example of mathematical notations used in the survey. B. Running a survey A survey created in Survnvote can be run for many times. Thus, a lecturer can use a survey or a quiz that he/she created in different class sections that he/she teaches. The results from each running of the survey can be accessed easily by the creator. This makes the creator does not need to go back to the creation page every time he/she needs to edit because of the incorrect display of the formula.

Fig.9. Result of individual question with different representation of data. C. Data representation There is various data representation provided by Survnvote. A comprehensive and informative data is presented during and/or after the survey finishes. The

Fig.7. Survey page

37

graph is presented in the form of a pie chart and a bar chart. Statistical data analysis can also be made available in the system. Thus for each question, the final result will be provided as a pie graph, a standard statistical data, a correlation table and also a cross tabulation table (Figure 9,10,11). Answer data and whole data of the final result can be exported to Excel and PowerPoint by clicking on the option.

system. When a respondent sends an SMS to answer a survey, the system will automatically create an account and send the SMS that contains username and password. By the time the respondent logins with the provided account, the phone number will be verified. This means that he/she now registered in the system with the details being synchronized. For the web based registration, a user needs to create an account and register her/his phone number. The system will send the request code to the phone. To verify the phone number, the respondent needs to confirm the phone number by copying the code to the verification page (please refer to Fig.12). Upon doing this, the user will then be registered in the system.

Fig.10. More Statistic options which are stored in excel format.

Fig.12. The verification of phone number IV.AN EXAMPLE OF USAGE This section provides a sample scenario of the usage of the Survnvote system in a lab-classroom setting. Using Survnvote a lecturer can pose questions in the class to get immediate feedback from the students. The questions can be a kind of question that will test students’ level of understanding about the topic that just delivered. This type of question will make the lecturer aware of how the students’ acceptance is in regard to the materials that he/she has just explained in the class. The Survnvote system has been used in the Network Programming class in Department of Computer Science, KICT, IIUM. After about 50 minutes of explaining in the class about I/O Multiplexing, the lecturer then pause for a while to pose questions to the students to get their level of understanding about the topic delivered. Students are asked to go to the URL of the questions and submit their answers. In about a minute after that, the whole class have submitted their responses, and the lecturer then finish the ‘survey’ and get the results.

Fig.11. The live survey updated on the PowerPoint slides D. User registration process To be a survey creator and a participant in certain survey, one should be registered as a user to the Survnvote system. There are three ways to register to this

The result in Figure 13 shows that the students have not grabbed the concept of the basic of I/O models, as 50% of

38

them gave the correct answer, while the other 50% went to the incorrect answer. Based on this result, the lecturer can respond by re-explaining the I/O concept to increase the understanding of the students, straight after the survey.

and quiz (surveyor gives answer before the survey is started). When it's used in the education environment, such as in a class-room, Survnvote can be used to evaluate students' understanding during a lecture by posing question(s) to the students and the lecturer can get immediate feedback and results. Students can vote using their mobile phone without any charge. Survnvote provides the users with an audience response system service and a web based survey application. Survnvote allowed a survey creator to be able to create their own survey questions and the respondent can participate in the survey using internet. Also a survey creator can create questions for getting instant responses from her/his audience, e.g. a lecturer can get an instant response from her/his students regarding the question(s) he/she posed in the classroom. In the class setting, Survnvote can be used to increase the understanding of student during learning process in the class. REFERENCES [1]

[2]

[3]

Fig.13. The Network Programming questions posed during the lecture

[4]

V. CONCLUSION This paper proposed some improved features of a web-based ARS which replaces the clicker with mobile phone and use the web instead of specialist software and infrastructure. Survnvote solved the main problem of traditional ARS which is high cost and need to physically install the receivers in the room. This system offers a more user friendly approach and improved features, e.g. the introduction of crowd management, Survnvote panel that easy to use, crowd group voting, and several ways to participate the vote.

[5]

[6]

[7]

[8]

Survnvote provides simple survey (single questionnaire), questionnaire (multiple questionnaires)

39

Poulis, J., Massen, C., Robens, E., Gilbert, M., “Physics lecturing with audience paced feedback”, American Journal of Physics, vol.66, issue 5, pp.439-441, 1998. Silliman, S.E. and McWilliams, L., “Observations on Benefits/Limitations of an Audience Response System”, Proceeding of the 2004 American Society for Engineering Education Annual Conference & exposition, 2004. Freeman, M., Bell, A., Comerton-Forde, C., Pickering, J. and Blayney, P., “Factors affecting educational innovation with in class electronic response systems”, Australasian Journal of Educational Technology, 23(2), 149-170, 2007. Fies, C. and Marshall, J., “The C3 framework: evaluating classroom response system interactions in University classrooms”, Journal of Science Education and Technology, 17(5), 483-499, 2008. Uhari, M., Renko, M., Soini, H., “Experiences of using an interactive audience response system in lectures”, BMC Medical Education, vol.3, 2003. Guthrie, R. W. and Carlin, A., “Waking the Dead: Using Interactive Technology to Engage Passive Listeners in the Classroom”, Proceedings of the Tenth Americas Conference on Information Systems, 2004. Ayu, M.A., Taylor, K., Mantoro, T., Active learning: Engaging Students in the Classroom Using Mobile Phones, Proceedings of ISIEA 2009, Kuala Lumpur, Malaysia, 2009. Maier, H.R., Student participation in lectures using mobile phones, Proceedings of the 20th Annual Conference of the Australasian Association for Engineering Education, Adelaide, Australia, 2009.


Semantic Query with Stemmer for Quran Documents Results Mohd Amin MohdYunus, Roziati Zainuddin and Noorhidawati Abdullah Faculty of Computer Science and Information Technology, University of Malaya, 50603, Kuala Lumpur, Malaysia. [email protected],{ roziati,noorhidawati}@um.edu.my

the data. In this paper, they describe QUICK, a novel system for helping users to construct semantic queries in a given domain. QUICK combines the convenience of keyword search with the expressivity of semantic queries. Users start with a keyword query and then are guided through a process of incremental refinement steps to specify the query intention [29]. A huge amount of web services are deployed on the Web, nowadays. These services can be used to fulfill online requests. Requests are getting more and more complicated over time. So, there exists a lot of frequent request that cannot be fulfilled using just one web service. For using web services, composing individual services to create the added-value composite web service to fulfill the user request is necessary in most cases. Web services can be composed manually but it is a too tedious and time consuming task. The ability of automatic web service composition to create a new composite web service is one of the key enabling features for the future for the semantic web. There are some successful methods for automatic web service composition, but the lack of standard, open, and lightweight test environment makes the comparison and evaluation of these composition methods impossible [30].

Abstract- The query-based on the result is lack of limited relevant documents in retrieval results. Therefore, query performance is considered to retrieve more relevant documents across language boundaries by applying semantic and stemmers which are more significant as stemming semantic query (SSQ). Therefore, this study is conducted with the purposes to investigate the integration semantic and stemmers approach against the queries and vice versa. Furthermore, it is also conducted to investigate the performance query based on total retrieve and relevant. The retrieval however, included the irrelevant documents because of the translation polysemy. Results from the experiments suggest that SSQ is most important process in cross language information retrieval (CLIR). It also found that semantic approach with stemmers contributes to better performance in retrieving more relevant and related Quran document results. Keywords- stemming; semantic; speech; query

I. INTRODUCTION Current text retrieval systems display generally relevant documents as a result to meet the query given in the process, and provide relevant results as relevant judgment. In the retrieval systems, ontology however [9][1] are most important in adding a semantic dimension between the user‟s query and the data sources. Some attempts for using ontology in search engines can be found in the literature [3]. In addition, there is not easier to find relevant domain ontology, and compile them [2]. Ontology can be also useful in result retrieval systems during query analysis [12][13]. Thus the system should be independent [16] in information delivery, from query to relevant result. The corpus is also distributed and heterogeneous, and results have to be integrated. The use of semantic technique to be embedded with ontology can be significantly to retrieve more relevant documents as a whole result.

III. BASIC FORMULA APPROACH Let W as total words which consists of word 1 (w1), word 2(w2), word 3 (w3) and the rest words (wn) in the search field. Thus the total words like this following formula

n

W  w w 1

where n is the last number of word and i is the first word. Therefore user can input words as many as they want as long as the total of retrieval results from the input words is not influenced. Regarding the result of the words, let D as total retrieval documents related to each word if in each document for the first word (w1d1, w1d2, w1d3….w1d6236), followed by (w2d1, w2d2, w2d3….w2d6236) and last word should be (wnd1, wnd2, wnd3….wnd6236). Therefore concluded result should be also like

II. RELATED WORK This research is motivated by the realisation that semantic technology can be used to develop computational tools in support of designers' creativity by focusing on the inspirational stage of design [28]. Constructing semantic queries is a demanding task for human users, as it requires mastering a query language as well as the schema which has been used for storing

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

(1)

40

D

t

W

d

translation is most important for those languages to investigate those information retrieval results. When the semantic query (SQ) is conflated with stemming algorithm (SSQ), this query can be translated and stemming algorithm according to the query keyed-in. Stemming algorithm removes the suffix, infix, and prefix of each word in the query to be root word in order to have more relevant documents in the results. The flow of each process is depicted in Figure 1.

(2)

d 1

where wn=W and should be any total depending on how many words to be key- in. The meaning of d is related to the verse or ayat or document which consists of the word given in the query.

Result retrieval

IV. EXPERIMENT APPROACH

User Interaction User Inquiry

These experiments have three languages which are the original and holly Quran in classical Arabic language in text, Malay Quranic documents collection [10] which is used by [25] as a domain in their research and English Quranic documents collection [14]. Each collection has 114 surahs and 6236 documents. Every document has its verse and chapter. All documents are as flat files in UTF-8, ASCII or EBCDIC text and searching process is through pattern matching [6]. The empirical research however generally has tested Malay query words are taken from Fatimah‟s collection as natural language queries [7] and the English as well as Arabic query words are translated from the Malay query words and query no. 27 has been tested and evaluated according to the formula as show in Table 1. Fatimah has obtained them by considering several guidelines put forward by [18] and [21]. Each query would be classified into keywords and replaced by target language. For example, there is Malay query, so it is called as source language and the target language is English or Arabic. Thus English as an example represents the translated word to retrieve English documents and if the query is English, the target language is Malay. The dictionary lists 1325 Arabic, Malay and English words in different flat files including 36 Malay query words selected. The query translation refers to the same index between Malay natural query language [7] and the translation of the natural query language in English or Arabic. When the keyword is Malay, then reference is to the English word at the same index or when the keyword is English and then reference is to the Malay word at the same index. Those words consist of synonym words also stemmed for looking their root words [7],[19],[23]. It is considering word by word in the text files.

Query Process Meaningful words Semantic dictionary files

Semantic Dictionary Look up Meaningful synonym words Stemming Meaningful stemming synonym words Retrieving Relevant documents Result

Figure 1. The Workflow of Cross Language Information Retrieval Based on Stemming Semantic Query

Then, searching process is done according to the type of process of results required. All documents are saved in “.txt” format file for UTF-8, ASCII or EBCDIC text. For searching process, word by word matching is used in the process. The matching words refer to the words similarity between query and documents in retrieving process. The query submitted to the system is also represented by translated query that is used to search the related files.

V. SYSTEM PROCEDURE The main interface is needed to get input from the user for retrieval. If the input is keywords, the results retrieved according to word by word results and redundant document names existed if merged. But querywords, retrieved according to the whole words as one at all and only when no redundant or unique document names retrieved rankly. Query translation can replace the origin query in to another language of the query. This

VI. RESULTS AND DISCUSSION The searching process is done according to the type of process of results required. All documents are saved in “.txt” format file for UTF-8, ASCII or EBCDIC text. For searching process, word by word matching is used in the process. The matching words refer to the words similarity between query and documents in

41

NATURAL LANGUAGE QUERIES

retrieving process. The query submitted to the system is also represented by translated query that is used to search the related files. All results of the query(ies) translation are referred to the natural language queries of Malay [7] and then translated into Arabic and English queries in this study. Every query is tested to evaluate each result which is matched with manual result as total relevant documents (TRE) for respective. The evaluation technique is used for precision and recall results [21]. Table 1 shows the formula to calculate the percentage of precision and recall. The SQT is quite suitable with the dictionary that consists of synonyms words and retrieves better results of the most relevant documents. It leads to help to search more and more documents in other languages. These examples also include stemming words after translating the word and matching the words in every document in collection in order to retrieve the most relevant document required from the query given. It is called as CLIR that focuses on search specific language if given query with the same language. Table 2 is referred from [25] thesis for query number 27 to prove that the semantic query in CLIR is significant to retrieve and provide more and more relevant and related results according to the available words provided in semantic files for three languages. Table 3 refers to the semantic query which is comparison between two testing input. Those are single query and semantic query and semantic stemming query through the process to display two different results according to specific language. In this context, those experiments are involved three languages which are Arabic, English and Malay. Table 4 refers to the retrieval results with the semantic query according to each language while table 5 refers to the retrieval results with SSQ according to each language. When the stemming semantic technique applied on query, the results are increasing as shown in Table 5 than Table 4. The significant difference shows the total retrieve (TRT) on Malay result at keyword (K), 568 to be 10128 and queryword (Q), 322 to be 951, then on English result at K, 303 to be 13627 and Q, 17 to be 5522 as well as on Arabic result at K, 6340 to be 6771, at Q, 431 to be 5274. It means that TRT is increasing after applying semantic method corresponds to the total retrieve and relevant (TRT).

Query No. 27

Malay

English

Arabic

Perkaitan nabi persamaan keturunan ciri-ciri rasul

Descendants of the prophet relevance equation characteristics messenger

‫عالقة‬ ‫األنبياء‬ ‫االصول‬ ‫نبي‬

TABLE 3 SEMANTIC STEMMING QUERY EVALUATION

TABLE 1 RECALL AND PRECISION FORMULA

TABLE 2

42

Language Malay

Q PERKAITAN NABI PERSAMAAN KETURUNAN CIRICIRI RASUL

English

DESCENDANTS PROPHET RELEVANCE EQUATION CHARACTERISTICS MESSENGER

SSQ PERKAITAN NABI PERSAMAAN KETURUNAN CIRICIRI RASUL ASOSIASI HUBUNG KAIT SIMBIOSIS UTUSAN ALLAH PESURUH ALLAH RASUL PERTEPATAN PERSERUPAAN ZURIAT DARAH PIUT CUCU SIFATSIFAT LELAKI UTUSAN ALLAH DESCENDANTS PROPHET RELEVANCE EQUATION CHARACTERISTICS MESSENGER CASSANDRA DRUID ASTROLOGER AUGUR CLAIRVOYANT DAYDREAMER DIVINER DREAMER ENTHUSIAST ESCAPIST FORECASTER ACCEPTED ACCUSTOMED ARRANGED AVERAGE BANAL BESETTING BOURGEOIS BUSINESSLIKE CENTRAL CHRONIC COMMON COMMONPLACE CONFORMABLE

‫عالقة األنبياء االصول نبي‬

Arabic

Language

‫عالقة األنبياء االصول نبي‬ ‫رسول شاعز هلهن قائد هلهن‬ ‫السوء نذيز الشؤم نذيز‬

TABLE 4 SEMANTIC QUERY EVALUATION TRE TRR Recall(%)

TRT K

Q

Malay

568

322

English

303

Arabic

6340

Precision(%)

K

Q

K

Q

K

Q

225

150

181

66.67

36.00

26.41

25.16

17

225

8

1

3.56

0.44

2.64

2.64

431

225

293

40

130.22

17.78

4.62

9.28

TABLE 5 SEMANTIC STEMMING QUERY EVALUATION TRT TRE TRR Recall(%) Precision(%)

Language K

Q

K

Q

K

Q

K

[2]

H. Baazaoui-Zghal, , M. A. Aufaure, and N. Mustapha Ben, “A modeldriven approach of ontological components for on-line semantic web information retrieval,” Journal on Web Engineering, Special Issue on Engineering the Semantic Web, Rinton Press, vol. 6, no. 4, pp. 309-336, 2007.

[3]

H. Baazaoui-Zghal, , M. A. Aufaure, and R. Soussi, “Towards an online semantic information retrieval system based on fuzzy ontologies,” Journal of Digital Information Management, vol. 6, no. 5, pp. 375-385, 2008.

[4]

T. d'Entremont and M. A.Storey, “Using a degree of interest model to facilitate ontology navigation,” vlhcc, pp. 127-131, 2009 IEEE Symposium on Visual Languages and Human-Centric Computing (VL/HCC), 2009.

[5]

T. Edward, “The visual display of quantitative information,” press, Chelshire, CT., 1983.

[6]

J. Elly, “The study of existing malay algorithm performed on words beginning with „D‟,” B.Sc. Thesis, Universiti Teknologi MARA, 2000.

[7]

A. Fatimah, “A Malay language document retrieval system an experiment approach and analysis,” Tesis Ijazah Doktor Falsafah Universiti Kebangsaan Malaysia, 1995.

[8]

S. Frintrop, E. Rome, and H. I Christensen, “Computational visual attention systems and their cognitive foundations: A survey,” ACM Trans. Appl. Percept., vol. 7, no 1, article 6, 2010.

[9]

T. Gruber, “Toward principles for the design of ontologies used for knowledge sharing,” International Journal of Human-Computer Studies, special issue on Formal Ontology in Conceptual Analysis and Knowledge Representation. Eds, N. Guarino and R. Poli, 1993.

Q

Malay

10128

951

225

583

147

259.11

65.33

5.76

15.46

English

13627

5522

225

563

213

250.22

94.67

4.13

3.86

Arabic

6771

5735

225

333

213

148

94.67

4.92

3.71

graphics

[10] H. Z. Hamidy and H. S. Fachruddin, Tafsir Quran. Translation. Klang, Klang Book Centre, 1987.

VII. CONCLUSION AND FUTURE WORK

[11] A. Katifori, C. Halatsis, G. Lepouras, C. Vassilakis, and E. Giannopoulou, “Ontology visualization methods,” A survey, ACM Comput. Surv., vol. 39, no. 4, article 10, 2007.

The stemming semantic results have a significant difference compared to single results. It has related documents to each another to be presented in the list of each result. The extension work will be focusing on hybrid stemming semantic query (HSSQ) results compared to SSQ. It is assumed that HSSQ is to be considered to retrieve more beneficial relevant retrieval results.

[12] V. Lopez, E. Motta, and V. Uren, “Poweraqua: fishing the semantic web,” Proceedings of the European Semantic Web Conference, ESWC 2006, Montenegro, 2006. [13] V. Lopez, V. Uren, E. Motta, and M.Pasin, “AquaLog: An ontologydriven question answering system for organizational semantic intranets,” Journal of Web Semantics, vol. 5, no. 2, pp. 72-105, Elsevier, 2007. [14] T. A. Muhammad and M. K. Muhammad, “Interpretation of the meaning of the noble Quran,” Dar-us-Salam Publications, 1999. http.//www.amazon.com/Noble-Quran-Interpretation-MeaningsLanguage/dp/996074079X.

ACKNOWLEDGMENT

[15] K. I Normaly, A. R. Nurazzah, and A. B. Zainab, “Terms visualization for Malay translated Quran documents,” In Proceedings of the International Conference on Electrical Engineering and Informatics, Institut Teknologi Bandung, Indonesia, pp. 17-19, 2007.

This research has been funded by the University of Malaya, under the grant number (PS210/2009B) and full-scholarship from the University of Malaya. Thus, I would like to forward our deepest thanks to Prof. Dr. Roziati Zainuddin and Dr. Noorhidawati Abdullah from the Faculty of Computer Science and Information Technology for their endless assistance, technical advice and co-operation.

[16] G. Pasi, “Flexible information retrieval: some research trends,” Mathware and Soft Computing, vol. IX, no. 9-1, 107-121, 2002. [17] H. Ping, X. X. Hua, and C. Ling, “Latent attribute Space Tree classifiers,” Journal of Software, vol. 20, no. 7, July 2009, pp. 1735−1745. Institute of Software, the Chinese Academy of Sciences.

REFERENCES [1]

[18] M. Popovic and P. Willett, “The effectiveness of stemming for naturallanguage access to Slovene textual data,” Journal Of The American Society For Information Science, vol. 43, no. 5, pp. 384-390, 1992.

M. A. Aufaure, B. Le Grand, M. Soto, and N. Bennacer, “Metadata- and ontology- based semantic web mining in web semantics and ontology,” D. Taniar & J. Wenny Rahayu eds., Idea Group Publishing, mars 2006, ISBN: 1591409055, 406 p, chapter 9, pp. 259-296, 2006.

[19] M. F. Porter, “An algorithm for suffix stripping,” Program, vol. 14, no. 3, pp. 130-137, 1980.

43

[20] L. Z. Qiang, C. H. Wu, X. B. Wen, , L. W. Qian, , W. J. Jia, and L. W. Jie, “A fast algorithm for synthesis of quantum reversible logic circuits,” Jisuanji Xuebao (Chinese Journal of Computers), Vol. 32, no. 7, pp. 12911303. July 2009. [21] G. Salton and M.J. Mcgill, “Introduction to modern information retrieval,” New York. Mcgraw-Hill, 1983. [22] G. Salton, “Experiments in automatic thesaurus construction for information retrieval,” In Proceedings Ifip Congress 1971, Ta-2, pp. 43-49, 1971. [23] K. Shereen, Arabic Stemmer, 2002. http://zeus.cs.pacificu.edu/shereen/. [24] C. Ware and P. Mitchell, “Visualizing graphs in three dimensions,” ACM Trans. Appl. Percpt., vol. 5, no. 1, article 2, January 2008. [25] M. A. M. Yunus, “Short query translation: A dictionary-based approach to cross language information retrieval,” Master of Computer Science, Thesis, Universiti Teknologi MARA, Malaysia, 2008. [26] R. J. R. Yusof, R. Zainuddin, M. S. Baba, and M. Z. Yusoff, “Visualization systems supporting the reading of Arabic document for non Arabic speakers,” Information Technology Journal, vol. 8, no. 1, pp. 16-27, 2009. [27] A. B. Zainab and A. R. Nurazzah, “Evaluating the effectiveness of thesaurus and stemming methods in retrieving Malay translated Al-Quran documents,” In Proceeding Of 6th International Conference On Asian Digital Libraries, pp. 653-662, 2003, Springer-verlag. [28] R. Setchi, Q. Tang and I. Stankov, "Semantic-based information retrieval in support of concept design," Advanced Engineering Informatics, vol. In Press, Corrected Proof. [29] G. Zenz, X. Zhou, E. Minack, W. Siberski and W. Nejdl, "From keywords to semantic queries--Incremental query construction on the semantic web," Web Semantics: Science, Services and Agents on the World Wide Web, vol. 7, pp. 166-176, 2009. [30] S. H. Yeganeh, J. Habibi, H. Rostami and H. Abolhassani, et al., "Semantic web service composition testbed," Computers & Electrical Engineering, vol. 36, pp. 805-817, 2010.

44


A Pilot Study in Using Web 2.0 to Aid Academic Writing Skills Azamjon Tulaboev, Alan Oxley Computer & Information Sciences Department Universiti Teknologi PETRONAS Bandar Seri Iskandar, 31750 Tronoh, Perak, Malaysia [email protected], [email protected]

Abstract – Today a wide range of Web 2.0 applications (such as Facebook, Twitter, Blogs, RSS, Podcasts and Wikis) are giving new challenges and opportunities in teaching and learning. The work described here concerns a pilot study to investigate the acceptability and effectiveness of using web 2.0 as an aid to learning. The work focuses on a university course on academic writing skills. The role of Web 2.0 applications are studied in building informal classes for traditional students and in adopting Web 2.0 practices by educators responsible for delivering existing courses. The extension of the UTAUT model is used as a research framework to evaluate the acceptability and effectiveness of web 2.0 applications.

technology advancements create new challenges and opportunities to the most enduring processes of teaching and learning. II. BACKGROUND OF STUDY Web 2.0 applications foster new modes of connectivity, communication, collaboration, sharing of information, content development and social organization [7]. The social characteristics of Web 2.0 increasingly feature in our daily life through social networks that are largely unbounded by space and time [8] allowing information to be shared, reported, researched and learned.

Key words - Experiential Pedagogy; Social Networks; Teaching and Learning; UTUAT model; Web 2.0 tools.

The usage of Web 2.0 technologies in higher education is still a new phenomenon and its integration into teaching and learning is in the initial phase [7]. Currently, with the social computing platforms of Web 2.0 being widely available [9], several Web 2.0 tools have emerged, and research is needed to determine pedagogical efficacy of these tools for teaching and learning.

I. INTRODUCTION Nowadays Web 2.0 applications (such as Facebook, Twitter, Blogs, RSS, Podcasts and Wikis) are enhancing work in academia. Whereas Web 2.0 has no complete explanation, it always refers to online interactions in which user groups both provide and receive content with the aim of collective intelligence [1, 2]. The Web can be seen as an ideal platform for enhancing challenging social expression, due to its ubiquity and openness [3]. Academic institutions at all levels are experimenting with these technologies to improve student learning experiences, and prepare them for a world in which work can be effectively accomplished through collaboration over the Internet, and geographic and time differences become increasingly irrelevant in sharing knowledge [2]. For instance, the collaboration tools of Facebook linked an entire generation in less than five years. CEO Mark Zuckerberg [4] recently announced that Facebook has 350 million active users – it can be compared to a population greater than that of the U.S.A. Mostly, users of today’s social networks are students of the “Net generation” [5], or as Prensky [6] calls them - “digital natives”. These are those who were born between 1982 and 1991. The Net generation and digital native students have different styles and expectations that require educators to reconsider pedagogical approaches [6]. Thus, Web 2.0

Recently conducted research at the crossroads of technology-enhanced learning (TEL) and the Web, focuses on adopting Web 2.0 tools such as tags for user modeling, personalization of mash-ups [10], and ontology and authoring [9]. TEL and Computer Mediated Communication (CMC) increases the amount and frequency of interaction between learners and educators [12]. Currently, through existing social networks, Web 2.0 facilitates the organization of informal classrooms for Net generation learners. The innovative character of new technology gives issues and opportunities, but not solutions [13]. Disk storage, collaboration tools and effective search logic now makes it technically possible to put all organizational knowledge online and to easily find the information needed [1]. However, tools alone have not created an effective learning environment [13]. Institutional experience of ICT usage [14] and students’ perspectives towards Web-based learning are becoming a considerable issue. As Grosseck [15] has claimed, certainly once engaged in using Web 2.0

Sponsorship by Universiti Teknologi PETRONAS

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

45

technologies, all the actors in the educational field will discover it worth the efforts and they will enjoy its benefits.

these new technologies [12]. Other arguments stated that the literature is rich in discussions on technology integration in the education process; very few studies elaborate on the effectiveness of the most recent Web-based tools from the student perspective [20]. These authors argued that the intention of using an interactive Web environment is not to replace classroom teaching but just to provide them more learning opportunities and to help them become active and autonomous learners.

Using Web 2.0 for learning is associated with such education term as ‘experiential learning’ and ‘studentcentered learning’. Web 2.0 should offer tremendous potential for learning as students already use it as part of their social life. The work described here build on learning web 2.0 researches. It is a limited attempt to explore the potential of Web 2.0 for teaching and learning. In particular, we wish to gauge the acceptability and effectiveness of using web 2.0 in the chosen setting. Furthermore, part of the research is an attempt to understand the forms of social interaction that contribute to learning.

Web-based tools can be a great assistance in language learning, however to get this assistance educators need to change their views regarding technology resources [11, 20]. Borau et al. [21] showed that the social, collaborative principles of Web 2.0 are reflected in its usage by language learners, if it used in an appropriate way. Virkus’s [7] research has confirmed that technology alone does not deliver educational success. It only becomes helpful in education if students and educators can do something useful with it.

The research is centered on an Academic Writing Skills course given at Universiti Teknologi Petronas. There are several lecturers assigned to the course teaching a large numbers of students. The lecturers have kindly agreed to allow web 2.0 practices to be used on the course on an experimental basis. It was arranged so that web 2.0 would only be used outside the classroom. A significant proportion of students’ time using various web 2.0 tools was to be spent engaged in social networks.

Critiques on using Web 2.0 in the learning process are a continuous issue of the usage of Web-based learning through a pedagogical perspective; relevant issues include students’ readiness and fluentness for the challenge of the knowledge society [15]. Also this author stated that when using Web 2.0 applications, we should be aware that abusive Web 2.0 can block or destroy information processing, and can decrease the quality of knowledge. It is a fact that the reliability and availability of publicly available Web 2.0 tools and services cannot be guaranteed [22] and are out of the control of the university.

III. LITERATURE REVIEW This section is intended to produce a theoretical basis for this study through existing literature. Adoption of Web-based learning has been assessed by several studies in different courses such as medical education [16], programming languages [17], language learning, distance education [18] and etc. Previous research shows positive experiences in Web-based learning, that it encourages students to be well-prepared and well-motivated students, and that the virtual classrooms are reasonably homogenous [19, 16]. According to Hwang et al. [17]: “several critical issues to be considered in programming courses, including the ways to motivate students’ interaction in or after class, methods to enrich students’ learning experiences, and facilities to assist students in sharing knowledge with their classmates.” Today, as Web technology has advanced it’s applications to the new platform of Web 2.0, there are great potential usage opportunities and challenges in the collective learning process [2]. Williams & Chin [14] studied how to support the active learning experience with using Web 2.0, and they gave a pedagogical strategy for today’s classrooms. They have researched the student’s and the instructor’s perspectives of usage of Web 2.0, in an effort towards increasing student engagement and Web 2.0 literacy. Web 2.0 tools and their increasing use in the learning process have presented educators with unique opportunities to further engage students in the learning environment using

Constructivism Constructivism is based on the argument that knowledge cannot be transmitted but has to be constructed by the individual [11]. Hence, teaching and learning is an active process of integrating information with pre-existing knowledge in a relevant context. Ullrich et al. [11] stated an argument that is inherent to pedagogy related to the use of technology, that is, the Web 2.0 pedagogy is best associated with constructivism and social learning. This claim is based on an analysis of the technological principles of Web 2.0 according to Alexander [23]. Hazari, North, & Moreland [9] suggest that constructivist theory be used to describe learning when using shared learning environments. In practice, many students have a high level of familiarity with certain tools but need guidance to explore them further, particularly in learning settings [14]. By building on familiar tools, learners became more creative and innovative in their exploration and adoption of available services. Subsequently Özgür & Özgür [20] stated that constructivism and technology are suited to a learning environment where learners may interact with each other in creating knowledge.

46

IV. METHODOLOGY

Research Process The purpose of this section is to develop an empirical research study and identify factors that are likely to influence the integration and adoption of Web 2.0 tools and services in the learning process. To achieve the objectives of the research, an experiment-based methodology was used in this study. The research scheme consisted of an implementation of blended learning design for student groups (Table 1).

In this part of the research the model and theory to be used will be specified. Recent literature supports several models that relate to technology applied in education, business and etc. Among them researchers prefer a model entitled the ‘Unified Theory of Acceptance and Use of Technology’ (UTAUT) [24] that can be extended for this research. The Extension of the UTAUT Model

The study was conducted during a regular academic semester. Three groups of students in English language courses (LBB 1042 - Academic Writing) were selected for experimental pedagogy, where two groups are the control group and the other is the experimental group. The experimental group was arranged into a blended learning community. Blended Learning Communities are groups which utilize face-to-face meetings as well as online meetings [7, 21, 26]. Supportive of the course syllabus for English, as part of our research we organized a project assignment that makes use of Web 2.0 tools such as Social Networks, Blogs, RSS, Wikis, Multimedia, Social Tagging, Instant Messaging. At the end of the semester, questionnaires were given to participants of this experiment. Descriptive statistics and SPSS tools were used for the data analysis part.

In support of the UTAUT model, Usluel & Mazmana [18] suggested using the model for further research into examining the adoption of Web 2.0 tools in learning. This is because UTAUT integrates the fragmented theory and research on individual acceptance of information technology, as well as an individual person’s perception and social influence. An important aspect of the UTAUT model is that it combines the essential elements of eight previously existing models: Diffusion of Innovation, Theory of Planned Action, Theory of Reasoned Action, Technology Acceptance Model I and II (TAM), Motivational Model, Social Cognitive theory, Model of PC Utilization [24]. Main suggestions for extension of the UTAUT model given by [24] that in terms of better understanding of technology adoption and usage behavior future research should go to determine and gauge additional boundary conditions of the model. Author [25] applied an extension f the UTAUT model in the context of business to business transactions on the web. And for this research previously conducted interviews’ results suggested additional variables to the UTAUT model such as barriers in using web 2.0 in education and different social subgroups formed based on ethnicity of an individual. Because of those suggestions research is going to attempt to identify and test additional as shown in figure 1. During our research this model will be tested in the context of acceptance of Web 2.0 tools in education.

Barriers

Table 1: Structure of Population

Ethnicity

Groups

Usage of Web 2.0 tools

Sample size

Duration of study

Experimental

Organized into a blended learning community

21

Weeks 1-14

Control 1

Free to use the tools but group not organized

25

Weeks 14-15

Control 2*

Free to use the tools but group not organized

19

Weeks 14-15

*where control group 2 has no overlap with experimental group

Performance Expectancy

Effort Expectancy

Behavioral Intention

Social Influence

V. RESULTS AND DISCUSSION

Actual Use of Web 2.0

To test the research model (Fig. 1) questions had been developed for each item through a comprehensive review of the literature on Web 2.0 and the technology acceptance theory. The range of measurement was formed as a fivepoint Likert Scale. To assess internal consistency and to make the scale items into a one dimensional scale, we used reliability analysis (covariance matrix method) employing Cronbach’s alpha reliability scores, where excluded items had an α value of less than 0.70 [27].

Facilitating Conditions

Gender

Age

User Experience

Voluntariness of Use

Fig. 1. Research Model (Adopted from Venkatesh et al., [24])

47

The data were analyzed using a Pearson productmoment correlation coefficient to find relationships among variables [Performance Expectance (PE), Effort Expectancy (EE), Social Influence (SI), Facilitating Conditions (FC), User Experience, Voluntariness of Use, Barriers, Behavioral Intention (BI), and Actual Use of Web 2.0 (AU)]. There were statistically significant positive correlations between the variables: AU and [PE, EE, SI, FC]; BI and [PE, SI, FC]; PE and [AU, BI, EE, SI, FC]; EE and [PE, AU]; SI and [PE, FC, BI, AU]; FC and [PE, SI, BI, AU] (Table 2). Here positive correlation means as one variable increases in value, the second variable also increases in value. These correlations are as expected interaction among items of the model. There was a statistically significant negative correlation between the variables: EE and Barriers, respectively p= -0.337. It means that students feeling easy to use web 2.0 tools increase (such as Blogs, Wikis, RSS, Social Networks, Multimedia and etc.) when barriers in using web 2.0 decrease or vice versa. There was no statistically significant correlation with User Experience (User Exp), Voluntariness of Use (Vol. use) and other items of the model (Table 3). Given results of the pilot study testing, it may not be surprising that results are fount little or no interaction between User Experience and Voluntariness of Use, it is possible due to students’ age range and their behavioral similarity toward the acceptability of web 2.0 tools in learning process.

a significant difference in the scores for Voluntariness of Use (Vol. use) and Effort Expectance (EE) (p = 0.029 and p = 0.027, respectively) (Table 5). Based on the mean of the test, it is very likely due to organize experimental group that students’ score of Voluntariness of Use seems more than control group where no engaged use of web 2.0 outside or inside a classroom. Similarly Effort Expectancy score where in experimental group students challenged to do something for course assignment and activities on the web 2.0 tools, they feel normal challenge toward the degree of ease of use the web 2.0 tools than control group where group free to use the tools but group not organized.

The next step in data analysis was an independentsamples t-test to compare variables of the research model in gender (Female and Male), ethnicity (Malaysian and International Students), age (under 20 and 21-22), and groups (experimental and control). For both Gender (Female and Male) and Age (under 20 and 21-22), there was no significant difference in the scores for AU, because p > 0.05. It would seem to have a surprise that the work found no difference of individual characteristics over intention to accept web 2.0 tools in learning process. Literature [28, 29] found that only in the lack of external motivators an individual characteristic to influence software adoption. Since web 2.0 tools are widely used in students’ social life, it indicates that students have external motivators to adopt web 2.0 tools in their individual learning. For Ethnicity (Malaysian and International Students), there was a significant difference in the scores for Social Influence (SI) and Performance Expectance (PE) (p = 0.011 and p = 0.031, respectively) (Table 4). These suggest that local and international students’ intention to adopt web 2.0 tools vary in terms of social influence and performance expectancy. The mean in an independent test (table 4) shows that Malaysian students have more concern about others believe regarding use of web 2.0 tools than International students and their individual believes that using the web 2.0 will help him or her to attain gains in study. For Groups (Experimental and Control 2), there was

This study has been conducted as a pilot study. To reach the proposed research objectives, we will endeavor to carry out a further experiment, learning from our experience with the pilot study. The main direction of our research will be to explore the acceptability and effectiveness of the use of Web 2.0 tools and services and their impact to the learning process in Higher Education. In future research we will define steps involved in implementing Web 2.0 tools in a practical classroom experiment and give ideas on the future of Web-based learning.

VI. CONCLUSION AND FUTURE DIRECTION The research applied the generally established theory of technology acceptance, originated in the MIS literature, to the issue of Web 2.0 technology use in Higher Education. The research model extended the UTAUT model. We chose UTAUT as the basis model because our experience suggests that the use of currently popular Web 2.0 technologies are reliant on performance expectancy, effort expectancy, facilitating conditions, and social influence. The study found several deviations from the classic MIS context, and the observed essential differences supported our research model, which was specific to the Web 2.0 context.

ACKNOWLEDGEMENTS The authors thank University Technology PETRONAS for providing a grant and facilities for the research.

48

Table 5: An Independent Samples Test (Between Groups)

Table 2: A Pearson Correlation Coefficients PE PE

EE

SI

FC

BARRIER

BI

AU

Std. Deviati on

1 .

EE

SI

.304(*)

1

.017

.

.463(**)

.068

EE

.591

.

.212

.576(**)

1

.024

.098

.000

.

.075

.337(**)

.233

.169

.579

.008

.073

.206

.

.343(**)

.066

.542(**)

.595(**)

.171

AU

N

1

.000

BARRIER

GROUPS Experimental group Control group 2

.292(*)

FC

BI

Vol. use

Experimental group Control group 2

Mean 16

4.0792

.44102

17

3.6941

.51697

20

3.5750

.65444

19

4.0526

.64323

Std. Error Mea n .110 26 .125 38 .146 34 .147 57

1 Levene's Test for Equality of Variances

1

.007

.604

.000

.000

.193

.

.293(*)

.399(**)

.397(**)

.560(**)

-.024

.185

1

.023

.001

.001

.000

.858

.148

.

F

t-test for Equality of Means Std. Sig. Mean Error (2Differe Differe tailed) nce nce

Sig .

t

.42 7

2.29 5

31

.029

.3850

2.30 6

30.719

.028

37

36.954

df

95% Confidence Interval of the Difference

* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed). Vol. use

Table 3: A Pearson Correlation Coefficients (User Experience, Voluntariness of Use and other items of model) User Exp User Exp Vol. use

Vol. use

1

.196

.

.144

PE .182 .161

BARRI ER

BI

.649

EE

EE

SI

FC

.193

.026

.148

-.246

-.031

AU .023

.127

.835

.250

.058

.810

.858

.155

-.224

.109

.253

.254

.104

.420

.058

.000

.196

1

.232

.222

.010

.144

.

.092

.097

.939

.99 0

2.29 7 2.29 8

Lower

Upper

.16779

.04284

.72726

.3850

.16696

.04440

.72570

.027

-.4776

.20792

.89891

-.05635

.027

-.4776

.20782

.89874

-.05652

* Correlation is significant at the 0.05 level (2-tailed). ** Correlation is significant at the 0.01 level (2-tailed).

Table 4: An Independent Samples Test (Ethnicity)

SI PE

Nationality Malaysian

N

Mean

53

3.5509

.62746

.08619

International

11

2.9818

.76134

.22955

Malaysian

50

3.7960

.56205

.07949

International

11

3.3818

.56889

.17153

Levene's Test for Equality of Variances

F

Std. Deviation

Sig.

t

t-test for Equality of Means Sig. Std. (2Mean Error taile Differe Differen d) nce ce

df

.828

Lower

Upper

2.639

62

.011

.5691

.21567

.13802

1.000 23

2.321

12.969

.037

.5691

.24520

.03927

1.098 98

2.208

59

.031

.4142

.18757

.03886

.7895 0

2.191

14.618

.045

.4142

.18905

.01032

.8180 5

PE .253

.617

[1]

O’Reilly, (2005). What is web 2.0: Design patterns and business models for the next generation of software. http://oreilly.com/pub/a/web2/archive/what-is-web20.html?page=1. [Accessed in October 15, 2009.]

[2]

T. Rienzo, and B. Han, Microsoft or Google web 2.0 tools for course management. Journal of Information Systems Education, vol. 20(2), pp. 123-127, 2009.

[3]

I. Rahwan, Mass argumentation and the semantic web. Journal of Web Semantics. vol. 6(1), pp. 29-37, 2008.

[4]

M. Zuckerberg, “An open letter from Facebook founder Mark Zuckerberg”. http://blog.facebook.com/blog.php?post=190423927130. [Accessed in December 2, 2009.]

[5]

D. G. Oberlinger, and J. L. Oberlinger, Educating the net generation. 2005. http://net.educause.edu/ir/library/pdf/pub7101.pdf. [Accessed in October 30, 2009.]

[6]

M. Prensky, Digital natives, digital immigrants. 2001. http://www.marcprensky.com/writing. [Accessed in November 3, 2009.]

[7]

S. Virkus, Use of web 2.0 technologies in LIS education: experiences at Tallinn University, Estonia. Program: Electronic Library and Information Systems. vol. 42(3), pp. 262-274, 2008.

95% Confidence Interval of the Difference

SI .048

REFERENCES

Std. Error Mean

49

[8]

H. Eijkman, Web 2.0 as a non-foundational network-centric learning space. Campus-Wide Information Systems. vol. 25 (2), pp. 93-104, 2008.

[23]

B. Alexander, Web 2.0: A new wave of innovation for teaching and learning? EDUCAUSE Review, vol. 41(2), pp. 32–44, March/April 2006.

[9]

S. Hazari, A. North, and D. Moreland, Investigating pedagogical value of wiki technology. Journal of Information Systems Education, vol. 20(2), pp. 187-198, 2009.

[24]

V. Venkatesh, M. G. Morris, G. B. Davis, and F. D. Davis, User acceptance of information technology: toward a unified view. MIS Quarterly, 27(3), pp. 425-478, 2003.

[10]

A. Oxley, Web 2.0 applications of geographic and geospatial information. Bulletin of ASIS, vol. 35(4), pp. 43-48, April/May 2009. http://www.asis.org/Bulletin/Apr-09/AprMay09_Oxley.pdf. [Accessed in August 10, 2009,]

[25]

Y. Benslimane, M. Plaisent, P. Bernard, “Using web systems for e-procurement: An extension of the unified theory of acceptance and use of technology”. ECIS 2004 Proceedings, Paper 8, 2004. http://aisel.aisnet.org/ecis2004/8

[11]

C. Ullrich, K. Borau, H. Luo, X. Tan, L. Shen, and R. Shen, Why web 2.0 is good for learning and for research: principles and prototypes. WWW 2008: proceedings of the 17th international conference on World Wide Web, pp. 707-714, 2008. http://www2008.org/papers/pdf/p705-ullrichA.pdf. [Accessed in September 24, 2009.]

[26]

R. Shrof, D. Vogel, An investigation on individual students’ perceptions of interest utilizing a blended learning approach. International Journal on E-Learning. vol 2, 279-294, 2010.

[27]

D. George, P. Mallery, SPSS for windows step by step. A simple guide and reference 16.0 update. Pearson Education Inc. Reliability Analysis. pp. 220-232, 2009.

[28]

M.B. Curtis, E. A. Payne, An examination of contextual factors and individual characteristics affecting technology implementation decisions in Auditing. International Journal of Accounting Information Systems, vol. 9 (2), pp.104-121, 2008.

[29]

T. Loraas, C. J. Wolfe. Why wait? Modeling the factors that influence the decision of when to learn a new use of technology. Journal of Information Systems, vol. 20 (2), pp. 1-23, 2006.

[12]

L. Li, and J. P. Pitts, Does it really matter? Using virtual office hours to enhance student-faculty interaction. Journal of Information Systems Education, vol. 20(2), pp. 175-185, 2009.

[13]

M. C. Colleen, “E-learning design 2.0: Emergence, connected networks and the creation of shared knowledge”. A Dissertation for the Degree Doctor of Philosophy, Capella University 2008.

[14]

J. Williams, and S. J. Chin, Using web 2.0 to support the active learning experience. Journal of Information Systems Education, vol. 20(2), pp. 165-174, 2009.

[15]

G. Grosseck, To use or not to use web 2.0 in higher education? World Conference on Educational Sciences 2009. Procedia Social and Behavioral Sciences 1, pp. 478–482, 2009.

[16]

V. Bernardo, M.P. Ramos, H. Plapler,… D.Sigulem, Web-based learning in undergraduate medical education: Development and assessment of an online course on experimental surgery. International Journal of Medical Informatics 73, pp. 731—742, 2004.

[17]

W. Y. Hwang, C. Y. Wang, G. J. Hwang, Y. M. Huang, S. Huang, A web-based programming learning environment to support cognitive development. Interacting with Computers 20, pp. 524– 534, 2008.

[18]

Y. K. Usluela, and S. G. Mazmana, Adoption of web 2.0 tools in distance education. World Conference on Educational Sciences 2009. Procedia Social and Behavioral Sciences 1, pp. 818–823, 2009.

[19]

P. Brusilovsky, J. Eklund, and E. Schwarz, Web-based education for all: a tool for development adaptive courseware. Computer Networks and ISDN Systems 30, pp. 291-300, 1998.

[20]

E. K. Özgür, and B.Özgür, Web 2.0 in learning English: the student perspective. World Conference on Educational Sciences 2009. Procedia Social and Behavioral Sciences 1, pp. 326–330, 2009.

[21]

K. Borau, C. Ullrich, J. Feng, and R. Shen, Microblogging for language learning: using twitter to train communicative and cultural competence. ICWL 2009, LNCS 5686, pp. 78–87, 2009. http://www.carstenullrich.net/pubs/Borau09Microblogging.pdf . [Accessed in October 15, 2009.]

[22]

C. D. Huang, and R. S. Behara, Outcome-driven experiential learning with web 2.0. Journal of Information Systems Education; Fall; 18(3), pp. 329-336, 2007.

50


Learning Acids and Bases Through Inquiry Based Website N. D. Abd Halim1, M. Bilal Ali2, J. Junaidi3 and N. Yahaya4

[email protected], [email protected], [email protected], [email protected] Educational Multimedia Department, Universiti Teknologi Malaysia, 81310 Skudai, Johor. of the content is delivered online. Typically, have no faceto-face meetings in the classroom. According to them this trend is seemed to increase from year to year and until 2008, where there are about 4.6 million students are moving towards online learning. One popular online application has been for educational use is web-based learning. Many studies have shown a web-based learning benefits and the potential to enhance teaching and learning process [18, 19, 20]. In web-based approach, students are able to choose how, when, and where they want to participate in the learning process [4]. In addition, in the term of accessibility the web-based application can be accessed anytime, anywhere and around the globe. Mean that, students can access the website as long as they have a computer together with internet connection. These are agreed by Neo et al. [21] where they said that popularity of the web base learning is due to the concept of learning “anywhere” and “anytime”. Else, other benefit in web-based learning is, the content is easily updated. As compared to CD-ROM application, it must be reduplicated and distributed again. But, through web based, we just need the developer to update files from a local computer to a server-computer. When students connect to the web for the next time, they will already have the latest version [22]. ICT has been integrated into teaching many subjects such as chemistry and others. Bayrak and Dori [23] found that integration of ICT in the process of teaching and learning chemistry can enhance students' knowledge of concepts, theory and chemical structure. Another study also found that ICT provides a positive impact on student achievement as providing a learning environment related to them [24]. By learning via website, students are provided with activities and an environment that allows them to participate actively in the learning process as well as assisted by teachers and peers [25].

Abstract- Chemistry is not an easy subject to learn. Many people regard chemistry as being too hard, too abstract, too mathematical, and only for very bright students. As a result, a negative attitude has developed about chemistry with students claiming chemistry is boring. Besides, most of the chemical concepts articulated by macroscopic, microscopic, and symbolic. This led to conflict and confusion in learning chemistry. Thus, the purpose of this project is to develop a web based learning material on Chemistry Form 4 based on Integrated Curriculum for Secondary Schools (KBSM) syllabuses. This website enables users to learn on their own about the topic of Acids and Bases. Constructivism theory and inquiry based learning approach were integrated in the development of this website. The main subtopics contain in the website are “Concept and Chemical Properties of Acids and Bases”, “Role of Water to Show the Properties of Acids and Alkalis” and “Strength of Acids and Alkalis”. This website was developed using Macromedia Dreamweaver 8 as the main platform whereas Macromedia Flash 8, Adobe Photoshop CS2 as well as Sound Forge 7 were used as supportive software. Finally, it is hoped that this website will become as a reference for students to learn this subtopic and to overcome their misconception about Acids and Bases. Other than that, this website is also useful for teachers and other people to get information more about Acids and Bases. I. INTRODUCTION

In the new millennium and the era of information technology, education field has moved rapidly towards the integration of technology specifically computers in the teaching and learning process. This is because computer has a great potential for enhancing teaching and learning outcomes. Michael [2] claims that, it is generally believed that ICTs can empower teachers and learners, promote change and foster the development of 21st century skills. According to Bayrak [3], computer based learning is becoming widespread and it also has been important method especially in teaching difficult subjects in science for over two decades. It is because by using computers, it can motivate students and also can enhance, extend or reinforce their learning in science [4]. Computer based learning is a method, which uses computer as a tool to give students strengthens, motivation and new experiences in gaining knowledge. It also gives opportunities to both students and teachers to learn and teach more quickly in order to achieve an active learning with computer technology [3]. Many researchers have proven the well-crafted use of computer in the learning process compare to traditional method in many disciplines such as Biology [5] Geometry [6], Science [7,8], Chemistry [9], Statistic [10], Sport Science [11], Nursing [12]. Nowadays, learning process is moving towards online application [13, 14, 15, 16]. Online learning as define by Allan and Seaman [17] is when the course where most or all

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

II. BACKGROUND OF PROBLEM

Chemistry is one of the branches of science that is important to learn because it enables students to understand the phenomena that occur around them [26]. However, chemical subjects closely related to abstract concepts, and this causes difficulties for students to learn [1].Students perception towards chemistry as it involves many abstract concept and it’s difficult to imagine. According to Taber [26], students need imagination and higher order thinking to learn and master the chemical concept. Understand the chemical concept is not only known what happened, but students also must know how to apply and explain it clearly and easily. These are the difficulties faced by students when they learn chemistry [1]. Chiu [27], states that several difficult topics in chemistry subject that cause misconception among students are

51

chemical equation, oxidation and reduction, electrolysis, mass and acid and base. Misconceptions can be defined as judgmental in view of the tentative nature of science and the fact that many of these conceptions have been useful to the students in the past [28]. Misconception also can be interpreted as one response in which the ideas are not developed align with the actual concept [29]. Many concepts in chemistry are related to each other. One of the basic concepts of chemistry and important to learn is acids and bases topic [32]. Many studies have shown that many students have difficulties in understanding the concept of acids and bases [28, 29, 30, 31, 32, 33, 34, 35, 36]. The study conducted by Cros et al. [31] shows that students are not able to explain concrete phenomena that occur behind the reaction of acids and bases. Students also fail to give examples of weak acids and bases and claim that pH scale as a tool for measuring the degree of acidity only. His research in the next two years in 1988, also states that students defined acid as a substance that has a pH less than 7. This is contrary to the definition given by the Arrhenius that acid is a release of hydrogen ions (H+) when the substance reacts with water. Ross and Munby [32] reported that students defined acid as a sour taste and produce heat. The difference between the theoretical definition of Bronsted and Arrhenius were misleading the students. The concept of donor and recipient of atoms was proposed by Bronsted-Lowry and the concept of donating and accepting hydrogen ions (H+),was used in the Arrhenius theory. Students found to be very difficult to see the continuity between these two theories in explaining the definition of acids and bases [34]. Hand and Treagust [33] has identified a number of misconceptions among students on the topic of acids and bases. They think that acid is a substance that only gives erosion of a material such as wood and iron and the process of neutralization is the process of removing acid. The study conducted by Barker [36], says that students like to explain the meaning of the acid using the word “eat” or corrosive materials without refer to the particle content. Although students know how to measure the pH and know the nature of corrosive acid, but students still find it difficult to relate the features and the properties of acids with the particles contain in the substance. Next, in 1993, a study was conducted by Nakhleh and Krajcik [35], obtained three key ideas in the concept of acids and bases among the students. First, the acid and base does not respond each other, but only form a physical mixture. Second, when acids and bases react together, they will stick together to form a particle. The third concept, the students felt that the reaction of hydrogen ions, (H+) and hydroxide ions, (OH-) is only suitable to form a soluble salt rather than the reaction of neutralization. The study conducted by Horton [28] at Arizona State University also has listed two misconceptions that often occur among students. They claimed that chemical reaction occurs when acid is added to the base. Second, students also defined the process of neutralization is the process of repealing the hydrogen ion (H+) and hydroxide ions (OH-). Chiu [29] in the study also proposed four students' misconceptions in this topic. First, students felt that the properties of the solution of acid and base were more or less the same. Second, students couldn’t explain the process of

neutralization clearly. For example, they said that when the sodium bicarbonate and acetic acid were mixed together, they became neutral. Third, students assumed when there is more hydrogen molecules in the acid, the stronger the nature of the acid and then the fourth, the students assumed that hydrogen molecules can be dissolved in the ionic solution only. Referring to all the misconceptions that have been listed, students will have difficulty in learning related topics, such as chemical equations and chemical reactions as described by [30]. In addition, apart from the chemical concept that is difficult to be understood, the way of the content being delivered by teachers can also influence the effectiveness of student understanding. Chemistry methods should undergo a paradigm shift from traditional methods to alternative approaches such as collaborative learning, self-learning, and problem-based learning through interactive multimedia, which involves students actively in the teaching and learning process. Demircioglu [30] suggested that teachers should use materials, aids and teaching strategies that encourage students to think actively, especially in helping students develop the concepts of acids and bases. Thus, in overcoming students' misconceptions and boredom of the acids and bases subtopics, the developing of multimedia application such as website can improve the quality of learning. Nakhleh and Krajcik [35] has suggested that the best way for teachers to teach chemistry, especially topic that involve abstract concept such as acids and bases topic is via the computer-based activities. This is evidenced by Idris et al. [37] which students had improved in achievement from pre to post test when they have learned acids and bases subtopic using multimedia software. III. OBJECTIVES OF STUDY

The objectives of this project is to develop an interactive acids and bases web sites based on form four Integrated Curriculum for Secondary Schools (KBSM) with the following features: i. Implementation of constructivism theory and inquiry based learning as an approach. ii. Incorporate multimedia elements such as text, graphics, audio, video and animation. IV. DEVELOPMENT PROCESS

The process of developing a website especially in the education requires a long period and systematic planning. This is to be done to ensure that the website is achieving the objectives that have been set. A developer must follow several steps as a guide in designing a website. For the purpose to develop this website, developer chooses the Hannafin & Peck (1998) model. Rationally, this model consists of three phases as well as easy to follow and the assessment phase can be done continuously. This model is simple and suitable for use in computerbased learning. This model consists of three phases, which are Need Assessment Phase, Design Phase and Development and Implementation Phase. Besides the three main phases in this model, there is another phase called Evaluation and Revision phase that is continuous to every phase of the three above. The evaluation process is an on going process. This

52

is to facilitate developer to improve website from time to time. In addition, by repeating the process of evaluation the developer can ensure that the website achieve the goal.

For the interface design phase, developer creates the interfaces that can connect between the users and computer. The interface consists of several elements such as background screen, panels, buttons, and multimedia elements such as text, graphics, animation, audio and video.

A. Need Assessment Phase This phase involves the analysis of user, the learning environment, the content of the topic, objectives to be achieved, and the goal of teaching. Analysis of users, especially students, is including their prior knowledge, gender, age, level of learning and learning style.

C. Development and Implementation Phase The development process is carried out with the help of programming system, authoring, graphics, audio, video, software and others. Software used to develop this website is Macromedia Dreamweaver 8. Else, the Macromedia Flash *, Adobe Photoshop CS2 and sound Forge 7 also have been used to develop animations, edit pictures and record sounds.

a. Analysis of user The target group for this website is form 4 and 5 students of secondary schools. However, other students can also use this website to obtain information of acids and bases subtopic. Before learn this topic, students have their previous knowledge about "Periodic Table of Elements" and "Chemical Bonding" which, in this subtopic, the students learn the characteristics of elements, compounds, and also the formation of the compounds.

V. THE WEBSITE OF ACIDS AND BASES

This website contains six menus such as Home, Learn, Quizzes, More Info, Help, and Web Master. This website is developed by applying the theory of constructivism and inquiry as the learning approach. In addition, to the use of text and diagrams, the use of multimedia elements such as animation, video and audio as well as a medium to convey the information for this topic acids and bases.

b. Analysis of content This phase is to identify the problems’ faced by students in the acid and base topic. The objectives to be achieved by students after using this website are : • State the meaning of acid and alkali according to Arhenius theory. • List the properties of acid and alkali compound. • Explain the role of water to show the properties of acid and alkali. • Explain the relationship between pH and the strength of acids and alkalis. • State the differences between strong acid and weak acid and the differences between strong alkali and weak alkali. In addition, this phase is needed to determine appropriate strategies and an approach that is align with the objectives to be achieved. Therefore, the application of constructivism as the learning theory with inquiry based learning approach is in place to achieve the objectives that have been targeted.

Figure 1 : “Learn” Interface

For each subtopic, learning process begins with a picture or a situation that raises questions and curiosity among user. This feature is applied in the "Have You Ever Wondered?" section. Subsequently, a number of hypotheses formed by users are shown in the section of "I Think That ..." . From the hypotheses that have been listed, students can do the investigation in the "Let's Investigate" section. Finally, after students completed their investigation, students can test what they have learned in "What Have I Learned?" section. To the right of each page, developer specifies the learning objectives that must be achieved by the students and the students’ prior knowledge. Besides, the developer provides a listing of science information in “Pit Stop "corner, so that user can see how acids and bases exist around them. Finally, there are the navigation buttons provided for the users such as forums, chat and e-mail. This navigation is applicable whether after or while they are in learning mode. This also enables users to interact with friends or other users.

B. Design Phase The phases include in the design phase are information design phase, interaction design phase and interface design phase. In the information design phase, the flow chart of the website was produced and the developer implements the theory and approach chosen for the site which is aligning with the learning objectives. In this context, the presentation of information or the content in this website is following the constructivism theory that has been listed by Kassim and Kamaruddin [38], Aris et al. [39], Valerine [40]. Meanwhile, the inquiry based learning is applied by the features that have been listed by Harwood [41] model. Next, at the interaction design phase, developer determines where and how the control will be given to the users. Else, developer also determines the route of exploration that will be taken by user to continue the exploration of a website. It is important to give user to choose freely. However, guidance should be provided so that users will not be lost. Therefore, developer needs to plan the interaction designs which are clear and easy to use.

VI. THE IMPLEMENTATION OF CONSTRUCTIVISM THEORY IN THE WEBSITE

This website implements the constructivism theory because this theory emphasizes student-cantered learning. Thus, developer implements the features of this theory that

53

has listed by Kassim and Kamaruddin [38], Aris et al. [39], Valerine [40]. Table 1 shows where the 12 constructivism characteristics were applied in the website.

The learning process for all three subtopics in this web site is according to the inquiry model that proposed by harwood [41]. He states that the characteristics of inquiry learning were make an observation on the problem given, forming questions, scan the existing knowledge, make a hypothesis, conduct an investigation, make a reflection on what they have learned and lastly share with others.

TABLE 1 THE IMPLEMENTATION OF CONSTRUCTIVISM THEORY IN THE WEBSITE Constructivism characteristic Encourage student with inquiry process through research or experiments

Encourage students to ask questions and dialogue with other students and teachers. Students have the opportunity to give any views on any concept. Students reflect on the learning process Teacher ask student to stimulate students to answer the question.

Use of collaboration and communication tools. Students can access the data or information from various perspectives. Students are freely to choose the topic that they want to learn by themselves. Students make a hypothesis.

Helping students to make the relationship between the topics before and the topic will be learn. Help students to relate the content with their life. Have a concept map that helps students to illustrate the whole topic.

Implementation in the website The "Let's Investigate" section allows students to investigate and find the answers to the problems given at the beginning. Investigation may be made by accessing the website, view video and animation. There are also the experiments that help students to find the answers and to develop a new concept. Chat room, forum and email are provided. So students can communicate with other students or teachers. This section also allows students to exchange ideas, share knowledge and make reflection after or during the learning process. The "Have You Ever Wondered?" section gives situations that stimulate students' curiosity and create enthusiasm among student. (see Figure 1) Chat room and forums allow students to collaborate (see figure 5) The websites, animation and video give various learning resources to students. Students are free to choose to any subtopic available on the website according to their needs. The "I Think That .." section help students to develop several hypotheses to a given problem. (see Figure3). Hypothesis formed will be investigated in the "Let's Investigate" section. "Content Relationship" help students to see the relationship between each subtopic they wanted to learn. The "Pit Stop" section provides the information about the relevance of science topics in their daily life.

A. “ Have You Ever Wondered?” section For the three subtopics in this website, the developer provides the difference situation for each subtopic. It can be image, animation or simulation that raises the question in students’ mind. For example, for the subtopic "Role of Water to Show The Properties of Acids and Alkalis", developer provides an animation of the changes of red litmus paper to blue colour when it was placed in a wet soap. After making an observation, the question “Why does the litmus paper turn from red to blue colour when it is in contact with wet soap?” was formed (see Figure 1). B. “I Think That..” section. "I Think That ..." is a section where the developer lists a number of possible hypotheses that students consider when making an observation on the images or situations given in the section before. Besides, students also had an opportunity to reflect on their prior knowledge in the "Flash Back" section. Four hypotheses are listed and students are required to guess which one is the correct answer. The "Which one is true" button will take students to a new page that allows them to do an investigation to prove their hypothesis. (see Figure 3)

Figure 3 : “I Think That..” section

"Content Mapping" is provided to help students to see the whole topic and also to get information. (see figure 2)

C. “Let’s Investigate” section This section helps students to find answers regarding the problem given by using a number of multimedia elements such as animation and video. Answers to the questions and hypotheses that were formed will be investigated and proved in this section. All the multimedia elements also help students to develop a new knowledge related to the topics studied. D. “What Have I Learned?” section After the learning process, students will test to choose several answers to the problems given before. There are several answers that have been listed. Students must select the correct number of the answers. In addition, students can also see whether the initial hypothesis is formed in the "I Think That .." section was rejected or accepted. Students can choose the answers by pressing the "Let's Check Your

Figure 2 : Content Mapping VII. THE IMPLEMENTATION OF INQUIRY APPROACH IN THE WEBSITE

54

Answer" button, then students need to enter the answer by typing the number in the space provided. (see Figure 4)

3 1 2 3

E. Chat and Forum section To enable students reflect on their learning, the developer has provided a forum site. Students can also discuss and share the knowledge with others. In addition, this forum is designed for students to provide feedback, comments or suggestions on the topics they have learned. What important is, via this forum the developer can see the effectiveness of student learning and see whether they are having a problem or not. Besides forum, chat rooms were also provided by the developer to allow students to collaborate with others while learning mode occurs. (see Figure 5)

4 5

VIII. EVALUATION OF THE WEBSITE

A. Evaluation by the design expert The evaluation for the web page design was carried out by three experts. The instrument used was a questionnaire Table 2 is the result of evaluation made by them.

1

The content presented is easy to understand Delivery of content and the information is well organized The software is easy for students to understand concepts related to the basic concepts of "Acids and Bases" The question given in the website easily understood by students The language is easy to understand Example given is realistic and clear Interaction design Users can control the speed of presentation of information in this website Users are not lost when exploring the website

2 3 4 5 6 1 2

1 Yes

3 Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes Yes Yes

Yes Yes Yes

Yes Yes Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes Yes Yes Yes Yes

Yes

Yes

Yes

Yes

Yes

Yes

C. Evaluation by the learning theory and learning strategy expert

TABLE 2 THE RESULT OF EVALUATION ON THE DESIGN OF THE WEBSITE Expert 2 Yes

Yes Yes No No Yes

The content experts were expert teachers for chemistry subject. They had more than ten years in teaching chemistry. For this evaluation purpose, developer interviewed three expert chemistry teachers. They said that the main advantages found in this website are the implementation of constructivism theory and inquiry approach where students are actively involved to develop their own knowledge. This is because, this theory is implemented student-centered learning, students learn via exploration, student learn independently and able to construct their own knowledge. For the inquiry purpose, the situation given at the beginning raises curiosity among students and the investigation process to prove the hypothesis are some of this inquiry features. The results of the interviews with experts also found that application of the theory and approach in this web site make it a new learning medium. In addition, forums and chat rooms provide opportunities for students and users to collaborate with others. Besides, it can be used to exchange ideas, suggestions, and reflection on the topics that had been learned. Hence, apart from the use of text and graphics, this website also integrates other multimedia elements such as animation, audio and video. Animation that was developed help students and users to understand the topics being taught. They also stated that the multimedia elements like this make students more fun and enjoyable to learn. However, the experts suggest to integrate the experiment by simulation in order to make students investigate by their own. This is because students can run the experiment and make it try and error in approving their hypothesis. This make students become more critical and creative thinker.

Figure 5 : Chat room

Information design

Yes Yes Yes Yes Yes

B. Evaluation by the content expert

Figure 4 : “What Have I Learned ” interface

No

Users easy to explore and get the required information Inteface design Attractive screen design Attractive and effective graphics Color is used effectively The combination of text, graphics, audio, animation enhance the learning The icons used are easily to understand and consistent

The expert of the constructivism theory was agreed that the website was implemented all the constructivism characteristic as listed in Table 1. For the evaluation for strategy in the website, Table 3 is the result made by the experts. TABLE 3 THE RESULT OF THE EVALUATION ON THE STRATEGIES IN THE WEBSITE No 1 2 3

55

Learning strategy The forum and chat room help students to understand more The objectives are stated clearly and can be achieved The approcoach selected is suitable with the topic

1

Expert 2

3

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

Yes

4 5

Users are encouraged to be creative and critical The delivery of content is suitable with the students learning style

Yes

Yes

Yes

Yes

Yes

Yes

[18] D. Clark, Hands-on investigation in Internet environments: Teaching thermal equilibrium. In: M.C. Linn, E.A. Davis, & P. Bell (Eds.), Internet environments for science education. Mahwah, NJ: Erlbaum. 2004. [19] M.C Linn, D. Clark, and J.D. Slotta, “WISE design for knowledge integration,” ScienceEducation, vol. 87, pp. 517–538, 2003. [20] M. Mistler-Jackson, and N.B. Songer, “Student motivation and Internet technology: Are studentsempowered to learn science?” Journal of Research in Science Education, vol. 37, pp. 459–479, 2000. [21] M. Neo, T.K. Neo and W.L. Yap, “Students' perceptions of an interactive multimedia-mediated web-based learning environment: A Malaysian perspective” in Ascilite Melbourne 2008 Conference, Melbourne, 2008. Retrieved July 8, 2010, from http://www.ascilite.org.au/conferences/melbourne08/procs/neo.pdf [22] K. Kevin, “Using the Web for Learning: Advantages and Disadvantages,” 2006. Retrieved July 5, from http://www.elearningguru.com/articles/art1_9.htm [23] M. Barak, and Y. J. Dori, “ Enhancing undergraduate students_ chemistry Understanding Through Project-Based learning In an IT Environment,” Journal of Science Education. vol. 89, no. 1, pp.117– 139, 2005 [24] Y. J Dori, M. Barak and N. Adir, “A Web-based chemistry course as a means to foster freshmen learning,” Journal of Chemical Education, vol. 80, no. 9, pp. 1084–1092, 2003. [25] M. Frailich, M. Kesner and A. Hofstein, “Enhancing students' understanding of the concept of chemical bonding by using activities provided on an interactive website,” Journal of Research in Science Teaching, vol. 46, no. 3, pp. 289 – 310, 2008. [26] K. S. Taber, Alternative Conceptions In Chemistry: Prevention, Diagnosis And Cure. London: The Royal Society of Chemistry. 2002. [27] M. Chiu, “A National Survey of Students Conceptions in Chemistry in Taiwan,” Chemical Education International, vol. 6, no. 1, 2005. [28] C. Horton, Student Preconceptions and Misconceptions in Chemistry. Integrated Physics and Chemistry Modeling Workshop. Arizona State University, June 2001. [29] Z. Ismail, S. N. Syed Idrus and M. A.Samsudin. Kaedah MengajarSains. Bentong: PTS Professional Publishing Sdn Bhd. 2006. [30] G Demircioglu., A. Ayas and H. Demircioglu, “Conceptual change achieved through a new teaching program on acids and bases,” Chemistry Education Research and Practice, vol.6, pp. 36-51, 2005. [31] D. Cros, M. Maurin, R. Amouroux, M. Chastrette, J. Leber, and M. Fayol, “Conceptions of First-Year University Students of The Constituents of Matter and The Notions of Acids and Bases,” European Journal of Science Education, vol. 8, pp. 305-313, 1986. [32] B. Ross, and H. Munby, “Concept Mapping and Misconceptions: A Study of High School Students’ Understandings of Acids and Bases,” International Journal of Science Education, vol. 13, pp. 11-23, 1991. [33] B. Hand and D. F. Treagust, “Student Achievement and Science Curriculum Development Using A Constructivist Framework,” Journal of School Science and Mathematics, vol. 91, pp. 172-176, 1991. [34] S. J. Hawkes, “Arhenius confuses students,” Journal of Chemical Education. vol. 69 no. 7, pp. 542 – 543, 1992. [35] M.B. Nakhleh, and J.S. Krajcik, “A Protocol Analysis of The Influence of Technology On Students Action Verbal Commentary & Though Process During a Performance of Acid and Base.,” Journal of Research in Science Technology. Vol. 30(a), pp. 1149-1168, 1993. [36] V. Barker, A Longitudinal Study of 16-18 Year Olds’ Understanding Of Basic Chemical Ideas. D.Phil. Thesis. Department of Educational Studies, University of New York. 1995. [37] N. Idris, E.Gnanamalar, S.Daniel, and R. Mohd Saat, Teknologi Dalam Pendidikan Sains & Matematik. Kuala Lumpur : Universiti Malaya. 2004. [38] A. H.Kassim and M. I. Kamaruddin, Ke Arah Pengajaran Sains dan Matematik Berkesan. Unpublished, Universiti Teknologi Malaysia. (2006). [39] B. Aris, R. Sumarni, and M. Subramaniam, Reka Bentuk Perisian Multimedia. Skudai : Penerbit Universiti Teknologi Malaysia, 2002. [40] N. M. Valerine, “Web-Based Learning and Instruction : A Constructivist Approach,” West Virgina Wesleyan College, 2000. Retrieved July 6, from http://www.idea-group.com/downloads/excerpts/IRM1931777047.pdf [41] W. S. Harwood, “A New Inquiry Model,” Journal of College Science Teaching. vol. 33, no 7. 2004.

IX. CONCLUSION

As a conclusion, hope that the developed website could be used as an alternative by students and teachers to study the topic "Acids and Bases". Hence, this website also expects to be one of the support teaching materials in schools today. This is because the website allows students to learn by themselves at any time and any where. In addition, the use of English as the medium can also help to improve students’ confidence as long as they learn using English in Science and Mathematics subjects. REFERENCES [1] G. Sirhan, “Learning difficulties in Chemistry: An Overview,” Journal of Turkish Science Education, vol. 4, no. 2, 2007. [2] T. Michael, Knowledge Maps: ICTs in Education, Washington, DC: infoDev / World Bank, 2005. [3] B. Bayrak, "To Compare The Effects Of Computer Based Learning And The Laboratory Based Learning On Students’ Achievement Regarding Electric Circuits," The Turkish Online Journal of Educational Technology, vol. 6, no. 1, 2007. [4] N. Abdul Ghani, N. Hamim and N. Ishak , Web-Based Learning In Science Education : Overview and Implementation For Primary School in Malaysia, International Conference on Education Universiti Brunei Darussalam. 2007. [5] B. B. Reed, “The Effects of Computer Assisted Instruction on Achievement and Attitudes of Underachievers in High School Biology,” Dissertation Abstracts International, vol. 47, no.4, pp. 1270-A, 1986. [6] L. P. Mccoy, “The Effect of Geometry Tool Software on High School Science Teaching, vol. 10, pp. 51-57, 1991. [7] M. Ibiş, The influence of Computer assisted Science Instruction on the success of the students, Ankara: Gazi University, Institute of Educational Sciences, 1999, Unpublished Master Thesis [8] C.Y. Chang, "Comparing the Impacts of a Problem-Based ComputerAssisted Instruction and the Direct-Interactive Teaching Method on Student Science Achievement, " Journal of Science Education and Technology, vol. 10, no. 2, 2001. [9] I. Morgil, S. Yavuz, Ö. Ö. Oskay and Seçil Arda, "Traditional and computer-assisted learning in teaching acids and bases," Chemistry Education Research and Practice, vol. 6, no. 1, pp. 52-63. 2005. [10] J. R. S. Fonseca, “On The Contribution of Using Computers In The Classroom In Teaching/Learning Statistics,” in 37th ASEE/IEEE Frontiers in Education Conference, Milwaukee, October 2007. [11] N. Vernadakis, E. Zetou, E. Tsitskari, M. Giannousi and E. Kioumourtzoglou, "Student attitude and learning outcomes of multimedia computer-assisted versus traditional instruction in basketball," Education and Information Technologies, vol. 13, no.3, pp. 167-183, 2008. [12] J. Bloomfield , J. Roberts, and A. While, "The effect of computerassisted learning versus conventional teaching methods on the acquisition and retention of handwashing theory and skills in prequalification nursing students: A randomised controlled trial," International Journal of Nursing Studies, vol. 47, no. 3, pp. 287-294, 2010. [13] K. W. White and B. H. Weight The Online Teaching Guide: A handbook of attitudes, strategies, and techniques for the virtual classroom. MA: Allyn & Bacon, 2000. [14] .S. M. Alessi and S. R. Trollip, Multimedia for Learning. MA:Allyn & Bacon, 2001 [15] Multimedia Development Corporation. The smart school roadmap 2005-2020: An educational odyssey,. 2005. Retrieved July 8, 2010, from http://www.msc.com.my/smartschool/downloads/roadmap.pdf. [16] S. S Liaw, H. M. Huang, and G. D. Chen, Surveying instructor and learner attitudes toward e-learning. Computers & Education, vol. 49, pp. 1066-1080, 2007. [17] A. Allen and J. Seaman, Learning on Demand Online Education in the United States, 2009, US : Babson Survey Research Group, 2010.

56


Using Genetic Algorithm To Break A Mono - Alphabetic Substitution Cipher S. S. Omran

A. S. Al-Khalid

College of Elec. & Electronic Techniques Foundation of Technical Education [email protected]

College of Elec. & Electronic Techniques Foundation of Technical Education [email protected]

Abstract- Genetic algorithms (GAs) are a class of optimization algorithms. GAs attempt to solve problems through modeling a simplified version of genetic processes. There are many problems for which a Genetic Algorithm approach is useful. It is, however, undetermined if cryptanalysis is such a problem. Therefore, this work trying to explore the use of Genetic Algorithms in cryptography. The focus is to be on substitution cipher. The principles used in this cipher form the foundation for many of the modern cryptosystems. The frequency analysis is used as an essential factor in objective function.

D. M. Al-Saady Foundation of Technical Education [email protected]

In the substitution ciphers the value of character or character string is changed when transforming the plaintext into ciphertext, but the position of the original string and its value replacement correspond exactly in the plain and ciphertext [4,7,8,9,10,11]. For example, if we encrypt the plaintext genetic algorithm in cryptography using a single character substitution cipher with a certain key, the ciphertext will be as shown in Fig.2. Cryptographic ciphers

I. INTRODUCTION The field of cryptology today represents that branch of information theory which deals with the security of information confidentiality. Methods in cryptology may be subdivided into two classes, namely that of cryptography (methods applied by authorized information sharers to design and develop encryption schemes in order to ensure confidentiality of information) and that of crypt-analysis (mathematical and statistical attempts by unauthorized persons to break cipher in order to reveal the meaning of the underlying protected data). The ciphertext is created by choosing a permutation of the 26-character alphabet and using it to replace each letter in the plaintext message [1]. A Symmetric cryptography ciphers may in fact be sub classified into block ciphers (in which blocks of data, known as plaintext, are transformed into ciphertext which appears unintelligible to unauthorized persons) and stream ciphers (which involve streams of typically binary operations and are well suited for efficient computer implementation). In this paper we shall focus on block cipher just substitution cipher as shown in Fig.1 [2,3,4]. In 1993 Spillman [5] for the first time presented a genetic algorithm approach to break a substitution cipher using genetic algorithm. He has explored the possibility of random type search to discover the key (or key space) for a simple substitution cipher. In this paper different parameters of the genetic algorithm were tested such as the population size and the time required finishing the algorithm for different number of generations. II. SUBSTITUTION CIPHER

Block ciphers

Substitution ciphers

Stream ciphers

Transposition ciphers

Product ciphers

Fig.1 Schematic representation of cryptographic cipher classification Text Character: a b c d e f g h i j k l m n o p q r s t u v w x y z

genetic algorithm in cryptography ( key cipher ) : H W U G C T V A F K D Y Q P B R JLFI X M SOZ N (cipher text)

VCPCIEUHYVBLYIAQEPULZRIBVLHRAZ

Is a symmetric cryptography ciphers, the substitution cipher is classified into two parts (Mono alphabetic and Poly alphabetic) [2,3,6].

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

Asymmetric Cryptographic ciphers

Symmetric Cryptographic ciphers

Fig.2 Example of a key of a single character substitution cipher

63

i. III. GENETIC ALGORITHMS ii.

A Genetic Algorithm is a general method of solving problems to which no satisfactory, obvious, solution exists. It is based on the idea of emulating the evolution of a species in nature and so the various components of the algorithm are roughly analogous to aspects of natural evolution [2,3,6,12,13,14]. The process begins by creating a random initial generation of individuals, sometimes called chromosomes that in some way represent the problem being solved. Pairs of members of the current population are selected and “mated” with each other by means of a crossover operation to produce members for the succeeding generation. Randomly selected members of the current generation also undergo mutation, in which random portions of “genetic” material are exchanged. The fittest members of each generation, as determined by a fitness function are then selected for the succeeding generation as shown in Fig.3. The crossover and mutation operations are controlled by the crossover rate and mutation rate parameters, which determine the proportion of the population that undergoes these changes [1].

iii.

5.

g. Go to 4. Output is the best solution. V. FITNESS MEASURE

The technique used to compare candidate keys is to compare statistics of the decrypted message with those of the language. The letter frequency is used for attacks against cryptographic cipher. The frequencies of letters are typically occur in natural languages are known and well documented [15, 16, 17]. Fig.4 shows the expected number of letters occurrences of letters in English language text of length 10000 characters. A natural choice as measure of fitness of a candidate key k for the cipher would be [3, 13, 18].

‫ܭܥ‬ൎ ߙ ෍ ห‫ݑ ܭ‬ሺ݅ሻ െ ‫ݑ ܦ‬ሺ݅ሻห൅ ߚ ෍ ቚ‫݅(ܾ ܭ‬ǡ݆) െ ‫݅(ܾ ܦ‬ǡ݆) ቚ ݅‫ܣא‬

Initialization

Evaluation

Selection

Apply crossover to get children (10 pairs). Apply mutation of 0.02% to the new children. Apply replacement to get a new population (20 keys).

൅ߛ ෍

௜ǡ௝ǡ௞‫א‬஺

Mutation

݅ǡ݆‫ܣא‬

ห‫ܭ‬ሺ௧௜ǡ௝ǡ௞ሻ െ ‫ܦ‬ሺ௧௜ǡ௝ǡ௞ሻหሺͳሻ

Here, A denotes the language alphabet (i.e., for English, [A . . . .Z]), K and D denotes known language statistics and decrypted message statistics, respectively, and the indices u, b and t denote the unigram, bigram and trigram statistics, respectively. The values of α, β and γ allow assigning of different weights to each of the three n-gram types [13].

Mating

Termination

Fig.3 The basic genetic algorithm IV. PROPOSED ALGORITHM The following is an outline of proposed algorithm: 1. Input the cipher text to the algorithm and relative character frequencies. 2. Initialize the algorithm parameters: maximum number of generations (M). 3. Generate the population p(0) keys randomly (for example 20 keys) each one with length of 26 letters. 4. For 1 to (M) do: a. Decrypt the cipher text by the 20 generated keys. b. Calculate the suitability of each key from every decrypted text using the formula of fitness. c. Sort the keys based on the increased fitness values. d. Keep 20% (2 pairs) of best fittest of p(0) for next generation. e. Use stochastically selection to choose 8 pairs from the 20 keys (parents). f. For 1 to 10 pairs do:

Fig.4 Relative frequency of letters in English text

VI. IMPLEMENTING THE ATTACK FOR MONO - ALPHABETIC SUBSTITUTION CIPHER

The attack is implemented by generating an initial candidate key pool p(0) of even cardinality, consisting of permutations of the set { a, b, c, . . . z }. The first generation is generated

64

randomly using a simple uniform random generator. Thereafter, the cipher text is decrypted using each permutation as a key, enabling us to assign a measure of fitness by using equation (1) to each candidate key. Pairs of candidate key are then stochastically selected for producing offspring after applying a method of crossover to each pair. The stochastic selection method is applied by choosing some pairs from the candidate which they have best fitness. Then Stochastic Universal Sampling works by making a single spin of the roulette wheel. This provides a starting position and the first selected individual. The selection process then proceeds by advancing all the way around the wheel in equal sized steps, where the step size is determined by the number of individuals to be selected. So if we are selecting n individuals we will advance by 1/n x 360 degrees for each selection. Note that this does not mean that every candidate on the wheel will be selected. Some weak individuals will have very thin slices of the wheel and these might be stepped over completely depending on the random starting position, as shown in Fig.5 [12].

in the random binary vector. The outstanding permutation entries are filled into child 1 in the order in which they occur in parent 2. Similarly, child 2 was formed by copying the entries (in the same positions) from parent 2 corresponding to zero entries in the random binary vector. The outstanding permutation entries are filled into child 2 in the order in which they occur in parent 1. This process is illustrated in Fig.7 [12, 19, 20]. After mutation has taken place the resulting set of candidate keys form the new pool population P(1). The crossover and mutation procedures are applied to this new key pool in order to produce the population P(2), and so forth, until some final population P(T) is reached (either after a prespecified number of generations, or when the minimum candidate key fitness exceeds some acceptable threshold).

P1

PKXAMLTIUBESJFG

HCORQYDVWNZ

P2

XRPUWZANMGOVSCQ TDFBJEHLKYI Crossover point

Intermediate child 1 PKXAMLTIUBESJFG

TDFBJEHLKYI

Intermediate child 2 XRPUWZANMGOVSCQ HCORQYDVWNZ Intermediate child 11 PKXAMLTIUBESJFG

□D□□□□H□□Y□

Crossover is the process of taking two parent solutions and producing from them a child. After the selection process, the population is enriched with better individuals. Crossover is a recombination operator that proceeds in three steps: i. The reproduction operator selects at random a pair of two individual keys for the mating. ii. A cross site is selected at random along the key length. iii. Finally, the position values are swapped between the two keys following the cross site. The traditional genetic algorithm uses single point crossover (which is used in this paper), when the two mating chromosomes are cut once at corresponding points and the sections after the cuts exchanged. Here, a cross site or crossover point is selected randomly along the length of the mated key and letters next to the cross sites are exchanged. If any of the exchanged chromosomes are already appears in the child, then these positions of chromosomes are left blank, then the letters that do not appear in a child are inserted in it. The outcome of this procedure is shown in Fig.6 [12, 18,19].

Intermediate child 22 XRPUWZANMGOVSCQ

H□□□□YD□□□□

After crossover, some keys are subjected to mutation. Mutation prevents the algorithm to be trapped in a local minimum. A random binary vector of the same length of the cipher key is generated and then using this vector to produce two offspring key from two parent candidate keys in the following way. Child 1 was formed by copying the entries (in the same position) from parent 1 corresponding to unit entries

1st Child

Fig.5 Stochastic selection

Child 1 PKXAMLTIUBESJFG

CDNOQRHVWYZ

Child 2 XRPUWZANMGOVSCQ HBEFIYDJKLT Fig.6 Applying crossover between two parents

Generating random binary number (26 binary numbers (0, 1))

110 0 0 1 0 10 1 0 0 0 0 1 1 1 1 1 0 1 0 0 0 1 0 First Parent (chosen randomly)

PRXAWLTIUGEVJFQHDOBCYMSKNZ Second Parent (chosen randomly)

XKPHMZANUBQVSFGTCORJEDLWYI

PRXKMLZIAGUVSFQHDOBTYCJENW 2nd Child

RXPHMTAGUEQVSFOBCYKJNDLWZI

Fig.7 Applying mutation for two parents

65

solution is for key 4 where the correct number of correct letters is 17. Bold lettering is used in table (1) for those letters in the population that appeared to be correct. The true key which is RNKIYUJEFCSZGOATMLPDWHVQBX is not in the final pool, this is due to the limited length of the ciphertext. Fig.9 shows the time required (elapsed) to finish the algorithm for different number of populations. The time required is increased as the number of populations is increased.

VII. RESULTS The attack to a mono-alphabetic substitution cipher was implemented for different number of populations. Fig.8 shows the relation between the fitness and number of generations, for population size of 20, 40, 60, and 80. It is clear from Fig.8 that the best fitness is reached after 300-400 generations. Table 1 shows the values of fitness for a 20 population (keys) and the number of correct letters obtained. It is clear that the best

No

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

100 200 300 400 500 600 700 800 900 1000

Population

Correct letter

Table 1: The best solution for fitness value and correct no. of letters when the population is 20 and after 400 generations

96 94 92 90 88 86 84

Fitness

fitness%

population size=20

95 89 94 94 85 94 94 89 89 90 91 88 89 87 89 88 88 89 89 89

RNCIYFJEBOSZDGAPMKUXLHVQTX RNKIYUHGTMJFVDAECLPZWOQSBX RNKIYBUELQHZWOASXFPDGTVCJM RNKIYJUELQHZGOASMFPDWTVCBX OSTIYBUELQKZWRAHXPFDGNVCJM RNTIYBUELQKZWOAHXPFDGSVCJM OHTIYBUELQNZWRASMPFDKGVCJX RHTIYBKELQAZWOGSXPFDVNUCJM RNTIYBKELQAZWOGSXPFDVHUCJM RNCIYFJEBOSZDGAPMKUXLHVQTW RNCIYBJEDGSZFKAOMLPTUHVQXW RNJVYULSFQKZICMTHEPDWOAGBX RNBIYUHGTMJFVDAECLPZWOQSKX RSJGYUKDLNXIWEAHMPFOVTZQBC RNSVYUJOLCGIWAEHMKPDZQTFBX AGKMWFJDLSCIUEYTNOPRZHVQBX RNSVYUEDPCJIWKLHMTFOZQGABX RNTIYJKELQAZWOGSXPFDVHUCBM ONTIYJKELQAZWRGSXPFDVHUCBM RNTIYBKELQAZGOWSXPFDVHUCJM

14 11 11 17 7 10 9 7 9 13 13 11 10 7 10 9 8 10 8 10

number of generations

fitness%

population size=40 91 90 89 88 87 86 85 84 100

200

300 400 500 600 700 number of generations

800

900 1000

fitness%

population size=60 90 88 86 84 82 80 78 100 200 300 400 500 600 700 800 900 1000 number of generations

VIII. CONCLUSION In this paper a genetic algorithm attack on a simple cryptographic cipher, called mono-alphabetic substitution, was implemented successfully. The algorithm was implemented using the MATLAB program. Different parameters were tested such as the number of population and the time required finishing the algorithm for different number of generations. It is apparent from the results that increasing the number of population above 20 was not helpful in retrieving the original key. This is evident in Fig.8 where the highest fitness was attained after 400 generations disregarding the number of the population. This can be

fitness%

population size=80 87 86 85 84 83 82 81 100

200

300 400 500 600 700 number of generation

800

900 1000

Fig.8 Fitness value for different values of population

66

explained by the great number of probable keys i.e. 26! that makes any population minor. Using GA to attack a mono-alphabetic substitution cipher proved to be an efficient method of cryptanalysis based on the aspect of comparing the frequency of letter occurrence in the model text.

REFERENCES [1]

[2]

[3]

elapsed time(sec)

population size=20 [4]

400 350 300 250 200 150 100 50 0

[5]

[6] 100 200 300 400 500 600 700 800 900 1000

[7]


[8]

elapsed time(sec)


1400 1200 1000 800 600 400 200 0

[10] [11] [12] 100 200 300 400 500 600 700 800 900 1000 [13]



elapsed time(sec)

2000 1500

[15]

1000

[16]

500 [17]

0

[18]

100 200 300 400 500 600 700 800 900 1000 number of generations

[19]

elapsed time(sec)

population size=80 3000 2500 2000 1500 1000 500 0

[20]

100 200 300 400 500 600 700 800 900 1000 number of genertaion

Fig.9 Elapsed time for different no. of population

67

M. Ralph & W.Ralph, A Word-Based Genetic Algorithm for Cryptanalysis of Short cryptograms, American Association for Artificial Intelligence (www.aaai.org). All rights reserved, pp.229233, 2003. D. Bethany , Genetic Algorithm in Cryptography , MSc. Thesis, Computer Engineering, Rochester Institute of Technology , Rochester, New York, July 2004. W. r. Grundlingh & jan, h. Van vuuren, Using Genetic Algorithms to break a simple cryptographic cipher, Retrieved March 31, 2003 from http://dip.sun.ac.za/`vuuren/abstract/ genetic.htm, submitted 2002. O. David, Evolutionary Algorithm for Decryption of Monoalphabetic Homophonic Substitution Ciphers Encoded as Constraint Satisfaction Problems, July 12-16, Atlanta, Georgia, USA, 2008. R. Spillman, M.Janssen, B. Nelson, & M. Kepner, Use of a genetic algorithm in the crypanalysis of simple substitution ciphers, Cryptologia 17(1), pp.31-44, January 1993. A. J. Clark, Optimization Heuristics for Cryptology, PhD, Thesis, Queensland University of Technology, February 1998. L. C. Washington, Introduction to cryptography with coding theory, Pearson Education, Inc., 2nd edition, 2006. A. K. Verma, Mayank Dave and, R. C. Joshi, Genetic Algorithm and Tabu Search Attack on the Mono-Alphabetic Substitution Cipher in Adhoc Networks, Journal of Computer Science 3 (3), pp.134-137, 2007. G. J. Simmons, Contemporary Cryptology, The Science of Information Integrity, The Institute of Electrical and Electronics Engineers, Inc., New York, 1991. D. Kahn, The Code breakers, The New American Library, Inc., USA, 1973. W. Stallings, cryptography and network security, principle and practices, Pearson Education, Inc., 4th edition, 2005. S.N.Sivanandam, S.N.Deepa, Introduction to Genetic Algorithms, Springer-Verlag Berlin Heidelberg 2008. T.Ragheb & A. Subbanagounder, Applying Genetic Algorithms for Searching Key-Space of Poly-alphabetic Substitution Ciphers, The International Arab Journal of Information Technology, Vol. 5, No. 1, pp.87-91, January 2008. R. L. Haupt & Sue Ellen Haupt, Practical Genetic Algorithms, John Wiley &Sons, Inc., 2nd Edition, New York, 2004. LD Callimahos & WF Friedman: Military cryptanalytics, part II, National Security Agency, Washington DC, 1956. LD Callimahos & WF Friedman: Military cryptanalytics, part I, National Security Agency, Washington DC, 1956. B Schneier: Applied Cryptography: Protocols, algorithms and source code in C, Jhon Wiley & Sons, Inc., New York, 1994. E. Pakize, O.Ali, T.Salih, Continuous Optimization Problem Solution With Simulated Annealing and Genetic Algorithm, 5th International Advanced Technologies Symposium (IATS'09), May 13-15, karrabul, Turkey, 2009. S. Tang, K.F. Man, S.Kwong and Q. HE, Genetic Algorithm and their Applications, ieee signal processing magazine, November, pp.22-37, 1996. D. F. Buthainah & A. A. Hamza, Enhanced Traveling Salesman Problem Solving by Genetic Algorithm Technique (TSPGA), World Academy of Science, Engineering and Technology 38, pp.296-302, 2008.


Control of a Single-link Flexible Manipulator Using Improved Bacterial Foraging Algorithm H. Supriyono, M. O. Tokhi, and B. A. Md. Zain Department of Automatic Control and Systems Engineering, The University of Sheffield, Sheffield, United Kingdom Email: [email protected] suppress vibration at the end point of the manipulator. In the controller design, the two objectives could be achieved by employing a hybrid control mediator comprising hub-angular control and vibration suppression control loop. Several hub-angular controllers have been proposed to control the SFM including proportional derivative (PD)-like control [5], hybrid control of neural fuzzy logic (FL) and genetic algorithm (GA) [6], modular neural network [7], and proportional integral derivative (PID) control [8]. The application of BFA in the control of SFM has not been reported yet, however it has been reported in a two-link rigidflexible manipulator [9], where hybrid BFA and particle swarm optimisation (PSO) are used for optimising hybrid fuzzy pre-compensated PD controller for trajectory control. It has been reported that a PD joint-based collocated (JBC) control strategy gives acceptable results in set-point tracking of a SFM [5]. However, heuristic tuning is used for the controller parameters. To improve the tuning process the BFA approach is adopted to optimise the controller parameters. The original BFA could be improved by using adaptable chemotactic step size regarding the nutrient value so that it could converge faster with better or comparable optimum nutrient value. By applying this mechanism, the chemotactic step size is big if the nutrient value is high and small if the nutrient value is low. Previous works done by researchers include using fuzzy (FL) [10], adaptive delta modulation principle [11], and simple linear function [12], [13]. Besides that, other adaptable chemotactic step size have been developed using three functions namely linear function, quadratic function, and exponential function [14]. The results in those works suggested that the application of adaptable chemotactic step size heads to faster convergence. In this work, the adaptable chemotactic step size mechanism using exponential function namely exponentially adaptive BFA (EABFA) [14] is adopted to optimise the PD JBC controller parameters for set-point tracking of a SFM system. A state space formulation model that represents the dynamics of the SFM derived from finite difference (FD) discretisation of governing dynamic equations of the system is used as a test bed. The rest of the paper is organised as follows: Section II discusses the model of SFM system used in the simulation and Section III presents the structure of EABFA-tuned JBC and the JBC-EABFA computation step. Simulation and discussion are presented in Section IV. Finally, Section V presents the conclusions drawn from the work.

Abstract-This paper presents an investigation into application of controller tuning using improved bacterial foraging algorithm (BFA) namely exponentially adaptive BFA (EABFA). The objective of the work is to evaluate the performance of EABFA in the controller tuning of a joint-based collocated (JBC) proportional-derivative (PD) control system. A simulation model of a single-link flexible manipulator system that incorporates hub inertia, structural damping and payload at the end-point of flexible arm is used as a test bed. JBC tuned by BFA is used to control the hub angular movement. An adaptable chemotactic step size mechanism that incorporates exponential function of the nutrient value is introduced to improve the original BFA. The developed EABFA has faster convergence with better or comparable optimum results than that of the original BFA. The performance of EABFA is assessed based on its convergence, optimum nutrient value and time-domain hub-angular response of the manipulator system. Keywords: Bacterial foraging algorithm; adaptable chemotactic step-size; joint-based control; flexible manipulator. I.

INTRODUCTION

Bacterial foraging algorithm (BFA) has been developed based on the foraging behaviour of E. Coli bacteria [1]. It has attracted significant attention from researchers, and has been used in several areas of control application such as design of multiple optimal power system stabilizers (PSS) [2] and optimisation of active power filter for load compensation [3]. It has been shown that BFA is able to find the optimum value and avoid being trapped in the local optima. In the optimisation process, besides its ability to locate the best optimum value the convergence speed of BFA also has to be considered. Because of its advantages [4] such as its lower energy consumption, smaller actuator requirement, safer operation due to reduced inertia, compliant structure, possible elimination of gearing, less bulky design, and low mounting strength and rigidity requirements, flexible manipulator systems have been are favoured in various applications over their rigid counterparts. However, the oscillatory behaviour of the flexible manipulator system during its operation needs special consideration. Generally, two dynamic behaviours to be controlled in a single-link flexible manipulator (SFM) system are the output trajectory (hub-angle) and vibration at the end-point of the flexible arm. Thus there are two control objectives: to control the hub-angular displacement and to

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

68

II.

By ignoring the effects of rotary inertia and shear deformation, a fourth order partial differential equation (PDE) representing the manipulator motion can be obtained as [5]:

FLEXIBLE MANIPULATOR MODEL

The flexible manipulator model considered in this work is derived from a laboratory-scale SFM that consists of three main parts: measuring instruments, motor, and a flexible arm. The measuring instruments consist of three sensors, i.e. a shaft encoder to measure hub angle displacement, a tachometer to measure hub velocity, and an accelerometer at the end-point of flexible arm to measure the end point acceleration rate. The arm of the manipulator is a beam made from aluminium material driven by printed-circuit armature motor at the hub. The outline of the SFM and its schematic representations considered in this work are depicted in Fig. 1 and Fig. 2 respectively [15]. The schematic representation of the SFM in Fig. 2 can be described as follows the stationary and moving coordinates are represented by POQ and ′ ′ respectively, is the hubangular displacement, is the applied torque at the hub by a drive motor, represents a payload mass, is the hub inertia, is inertia associated with the payload. The physical parameters of the flexible arm used in this work are as follow: length = 960 mm, width = 19.008 mm, thickness = , 3.2004 mm, mass density per unit volume = 2710 the second moment of inertia = 5.1924 10 , the , moment of inertia = young modulus = 71 10 , and = 5.86 10 . In this work, the 0.04862 impact of gravity is neglected. A pinned-free flexible beam configuration can be used to model the flexible manipulator system with inertia at the hub and payload mass at the endpoint. The Lagrange equation and modal expansion method [16] are used to model the system. Shaft encoder

l

Hub

Ih

(1) 0 is the resistance to strain velocity. The FD method is where used to solve the motion equation and to develop a suitable simulation environment characterising the behaviour of the system. The FD method is chosen in contrast to finite element (FE) method because it has less computational complexity and could model the dynamics of the flexible manipulator adequately. The state space formulation of dynamic motion representation using FD method developed in [5] is used through this work. ,

Flexible arm

,

III. JBC CONTROL TUNED BY EABFA A. JBC-EABFA controller structure The JBC PD control for the flexible manipulator can be formulated as: (2) where is the control command, is proportional gain, is the derivative gain, is the error ( is the reference angular displacement), and is the actual angular displacement. Here, EABFA is used to find optimum values of JBC PD control parameters as shown in the block diagram in Fig. 3. The nutrient media which will be optimised by EABFA is the cost function formulated based on the hub-angle error. A bang-bang signal with 75 degree amplitude which consists of one positive pulse and one negative pulse is used as reference input. In the simulations to be carried out with this reference signal, overshoot response during the positive pulse and undershoot with corresponding setting and corresponding settling of system response during the negative pulse will be of interest to minimise. In order to suppress excessive overshoot and undershoot in the output, maximum undershoot is incorporated in the cost function together with hub-angular error. Thus the cost function considered includes the mean squared error (MSE) and absolute of maximum undershoot, and this is be formulated as:

w

E , I an d ρ

Motor

,

Mp

Tachometer

Figure 1. Outline of the flexible manipulator system [15]

∑

|

|

(3)

where is the angular displacement error, is the total number of data points, is and are weighting factor. The maximum undershoot, bigger value of will result bigger suppression on overshoot and undershoot. The incorporation of the absolute maximum undershoot in the cost function may lead to slowing the response of the flexible manipulator.

Figure 2. A schematic representation of the flexible manipulator system [15]

69

value, the global minimum solution should be non-negative. The EABFA is not valid if the global minimum solution is negative. Thus, when the nutrient value is big, it means the bacteria’s position is still far away from the global optimum position so that big step size is needed to head the place with nutrient while when the nutrient value is small it means that bacteria’s position is close enough to the global optimum position so that small step size is needed so that bacteria are able to reach the global optimum position. With this new chemotactic step size, to find the place with high nutrient value, bacteria will apply random walk with adaptable step size depending on the nutrient value. Thus, the computation steps of EABFA to find optimum JBC controller parameters can be formulated based on the work from [1] with modification on the chemotactic step size as follows:

Figure 3. EABFA-tuned JBC controller of flexible manipulator

The global optimum value of is equal to zero. In terms of BFA computation, lower value means higher nutrient level so that the highest nutrient level achieved when equals to zero. Thus the main aim of optimisation in this work is to find minimum value of cost function in equation (3). Smallest value of means the output of the flexible manipulator is closest to the reference input.

1. 2. 3.

B. JBC-EABFA computation The standard BFA (SBFA) is developed based on the foraging strategies of E. Coli where bacteria move to the place where high nutrient exist and avoid noxious places called chemotaxis [1]. From the optimisation point of view, suppose that it is desired to find the minimum of , , where there is no measurement or there is no analytical description of the gradient , ideas from bacterial foraging mechanism can be used to solve this non-gradient optimisation problem. Suppose that is the position of a bacterium and represents the combined effects of attractants and repellants from the environment, with 0, 0, and 0 denoting that the bacterium at location is in nutrient-rich, neutral, and noxious environments, respectively. Then, E. Coli bacteria will try to climb up the nutrient concentration (find lower and lower values of ), avoid noxious substances (avoid being at positions where 0), and search for ways out of neutral media (the positions where 0) optimally by implementing a type of biased random walk. With SBFA [1], to find places with high nutrient level, bacteria use random walk with certain constant value for whole computational process regardless of the nutrient value. In the current work, the chemotactic step size is made adaptable regarding the nutrient value of the bacteria’s current position using exponential function of the nutrient media. Thus, in EABFA [14] the chemotactic step size for every bacterium can be formulated as:

4. 5.

, , ,

(5) 1

|

Elimination-dispersal loop: for 1,2, … , , do 1 , do 1 Reproduction loop: for 1,2, … , 1 Chemotaxis loop: for 1,2, … , , do a. For 1,2,3, … , , run a chemotactic step for bacterium : b. Compute the nutrient value of every bacterium ( , , , ) with in equation (3) used as the nutrient function. , , , to save this value since a better cost via a c. Put run may be found. with each element d. Tumble: Generate a random vector ∆ , 1,2, … , , a random number in 1,1 . ∆ e. Move: compute the position of every bacterium ∆ 1, , , , ∆ ∆ is exponentially adaptable chemotactic step size. f. Compute the nutrient value of every bacterium ( , , , ) with in equation (3) used as the nutrient function. g. Swim: i. Put 0 (counter for swim length) (if have not climbed down too long) ii. While • Count 1 (if doing better), then • If , 1, , , 1, , and calculate the position of every bacterium ∆ 1, , 1, , ∆ ∆ is exponentially adaptable chemotactic step size. Use this 1, , to compute , 1, , ) as in sub step f above. . • Else, h. Go to next bacterium 1 if (i.e., go to sub step b above) to process the next bacterium. , go to step 3. If Reproduction: a. For the given and , and for each 1,2,3, … , , let

|

is exponential adaptable chemotactic step size where for every bacterium, is tune-able maximum step size, is tune-able positive factor, and are tune-able scaling factors, and is the nutrient value for every bacterium. The requirement that has to be met using this adaptable chemotactic step size is, as it incorporates absolute nutrient

6. 7.

8.

70

Be the health of bacterium . Sort bacteria and chemotactic (higher cost parameters in order of ascending cost means lower health). values die and the other b. The bacteria with the highest bacteria with the best values split (and the copies that are made are placed at the same location as their parent). , go to step 2. If , Elimination-dispersal: for 1,2,3, … , , with probability eliminate and disperse each bacterium (this keeps the number of bacteria in the population constant). , then go to step 1; otherwise end. If

IV.

1.1266 10 , than JBC-SBFA, i.e. 1.1855 10 . The lower value resulted by JBC-EABFA results in better time-domain performance, i.e. shorter rise-time, decline-time and settlingtime both over the positive and negative pulses than those of JBC-SBFA. Both controllers are also able to suppress unwanted overshoot and undershoot. Fig. 6 presents the timedomain hub-angular displacement output with the controllers.

RESULTS AND DISCUSSIONS

In this work, first of all, the SFM is simulated in the open loop condition to characterise its hub-angular displacement dynamics without and with payload at the end-point of the flexible arm. Then the JBC-EABFA is applied to control the SFM. The time-domain hub-angle output is used as a basis of the performance comparison. All simulations were carried out using Matlab/Simulink software. A. Behaviour in open-loop To characterise its basic response regarding the incorporation of payload at the end-point of flexible arm, an open loop simulation is performed. Various payload values, i.e. no payload, 30 gr, and 50 gr are applied at the end-point of flexible arm. The simulation results depicted in the Fig. 6 show that the open-loop response the addition of payload at the end-point of the flexible arm makes the SFM’s response slower and also reduce the hub-angle output. Bigger payload produces slower response and smaller hub-angle output. It can be noted from Fig. 4 that the hub-angle output is oscillatory around 51 degree, 32 degree, and 27 degree for no payload, 30 gr payload, and 50 gr payload respectively.

Figure 4. Open-loop hub-angular displacement of flexible manipulator with various payloads.

B. Simulation with JBC-EABFA controller In order to evaluate its effectiveness, the performance of JBC-EABFA is compared with JBC control tuned by standard BFA (JBC-SBFA). The comparisons are made based on their convergence, optimum value of the cost function that can be achieved and the time-domain performance of hub-angular displacement with various payloads. In the simulation, the same initial parameters as: =2, =4, =16, =2, =3, = /2, =3, and =0.25 were used for both SBFA and EABFA and the initial positions of the bacteria were selected randomly across the space of cost function. A structural damping factor equal of 0.024 was used in the simulation of SFM. In the convergence analysis, the nutrient value is plotted against the total steps of BFA, calculated as the total steps through the entire the computation, i.e. . Thus the nutrient value in every step means the position of bacterium that has the smallest cost function value. In the time-domain representation, the hub-angle output is plotted against the time. In this work, the performance of the controller is assessed based on 8 parameters of the timedomain hub-angle output as: the maximum overshoot ( ) over the positive pulse, the rise time ( ) over the positive pulse, the settling time ( ) over the positive pulse, the steady state error ( ) over positive pulse, the maximum undershoot ( ) over negative pulse, the decline time ( ) over the negative pulse, the settling time ( ) over the negative pulse, and the steady state error ( ) over the negative pulse. The convergence of JBC-SBFA and JBC-EABFA for no payload attached at the end-point of the flexible arm as depicted in Fig. 5 show that JBC-EABFA converged significantly faster than JBC-SBFA. For the resulted optimum value, the numerical results presented in Table I show that JBC-EABFA achieved slightly better optimum value, i.e.

Figure 5. Convergence of the two controllers without payload

71

Figure 6. Time-domain hub-angle output plot with the two controllers without payload

Figure 8. Time-domain hub-angle output with the two controllers with 30 gr payload.

With 30 gr payload incorporated at the end-point of the flexible arm, the convergence plot in Fig. 7 show that JBCEABFA was able to converge significantly faster than JBCSBFA. The numerical results for optimum nutrient value highlighted in Table II show that JBC-EABFA has achieved slightly smaller value, 1.33 10 , compared to 1.3603 10 for JBC-SBFA. For the time-domain results, JBC-EABFA has achieved shorter rise-time and decline-time but has longer settling time both over the positive and negative pulses. Compared to the result without payload, the responses with both JBC-SBFA and JBC-EABFA with 30 gr payload were slightly slower. From the steady-state value, it can be noted that both JBC-SBFA and JBC-EABFA were able to recover the hub-angle output decrement due to payload attachment at the end-point of the flexible arm. Also, both controllers were able to suppress the unwanted overshoot and undershoot. The time-domain performances of both JBC-SBFA and JBCEABFA are depicted in Fig. 8.

Similar to the case of no payload and 30 gr payload, JBCEABFA with 50 gr payload was able to converge faster than JBC-SBFA. The convergence plot for the controllers is depicted in Fig. 9. The numerical results in Table III show that JBC-EABFA resulted in lower value, 1.4792 10 , compared to 1.481 10 of JBC-SBFA. For the time-domain performance, JBC-EABFA also had slightly shorter rise-time, decline-time, and settling-time than JBC-SBFA. The timedomain hub-angle outputs with JBC-SBFA and JBC-EABFA with 50 gr payload were slightly slower than that without payload and with 30 gr payload. The steady-state error values show that both controllers were able to recover the hub-angle output decrement caused by the payload attachment at the endpoint of flexible arm. Also both controllers were able to suppress the unwanted overshoot and undershoot in the output. The time-domain hub-angle output plot for the two controllers is depicted in Fig. 10.

Figure 9. Convergence of the two controllers with 50 gr payload. Figure 7.

Convergence of the two controllers with 30 gr payload.

72

Figure 10. Time-domain hub-angle output with the two controllers with 50 gr payload. V.

CONCLUSION

In this work, an improved BFA, namely EABFA, has been adopted to tune the controller parameters in JBC control structure for set point tracking of a SFM application. Based on the convergence plot, optimum nutrient value, and timedomain hub-angle output, the simulation results suggested that the JBC-EABFA has better performance than that of JBCSBFA. Also, both JBC-SBFA and JBC-EABFA were able to track the reference hub-angle with various payloads attachment at the end-point of the flexible arm. Future study will investigate strategies for vibration suppression at the endpoint of the flexible arm. TABLE I. NUMERICAL RESULTS OF CONTROLLER’S PERFORMANCE IN THE TIME-DOMAIN (NO PAYLOAD) Controller Optimum (s) (s) (s) (s) JBC-SBFA 0.9394 0.4887 0.8432 1.4370 1.1167 1.7290 1.1855 10 JBC-EABFA 0.6957 1.2703 0.8877 1.5004 1.8297 0.7325 1.1266 10 TABLE II. NUMERICAL RESULTS OF CONTROLLER’S PERFORMANCE IN THE TIME-DOMAIN (PAYLOAD 30 GR) Controller Optimum (s) (s) (s) (s) JBC-SBFA 0.8942 1.1204 1.1797 1.3518 1.2397 0.6658 1.3603 10 JBC-EABFA 0.8781 1.5438 1.0951 1.8068 2.1481 1.0146 1.33 10 TABLE III. NUMERICAL RESULTS OF CONTROLLER’S PERFORMANCE IN THE TIME-DOMAIN (PAYLOAD 50 GR) Controller Optimum (s) (s) (s) (s) JBC-SBFA 0.9921 1.3433 1.3124 1.5810 1.3750 0.7971 1.481 10 JBC-EABFA 0.9803 1.2317 1.2904 1.4719 1.3498 0.7756 1.4792 10 ACKNOWLEDGMENT

[8]

Heru Supriyono acknowledges the financial support of National Education Department of Republic of Indonesia, and Muhammadiyah University of Surakarta (UMS), Indonesia.

[9]

[10]

REFERENCES [1] [2]

[3] [4]

[5] [6] [7]

K. M. Passino, “Biomimicry of bacterial foraging for distributed optimization and control”, IEEE Control Systems Magazine, June 2002, pp. 52-67. T. K. Das, G. K. Venayagamoorthy, and U. O. Aliyu, “Bio-Inspired Algorithms for the Design of Multiple Optimal Power System Stabilizers: SPPSO and BFA”, IEEE Transactions On Industry Applications, vol. 44 no. 5, 2008, pp. 1445-1457. S. Mishra, and C. N. Bhende, “Bacterial foraging technique-based optimized active power filter for load compensation”, IEEE Transactions on Power Delivery vol. 22 no. 1, 2007, pp. 457-465. W. J. Book and M. Majette,. “Controller design for flexible distributed parameter mechanical arms via combined state-space and frequency domain techniques”, Transaction of ASME Journal of Dynamic Systems, Measurement and Control, vol. 105, no. 4, 1983, pp. 245-254. H. Poerwanto, “Dynamic simulation and control of flexible manipulator systems”, PhD thesis, Department of Automatic Control and Systems Engineering, The University of Sheffield, UK, 1998. M. N. H. Siddique and M. O. Tokhi, “GA-based neural fuzzy control of flexible-link manipulators”, Engineering Letters, vol. 13 no. 2, 2006. Sharma, S. K., Irwin, G. W., Tokhi, M. O., and McLoone, S. F., “Learning soft computing control strategies in a modular neural network architecture”, Engineering Applications of Artificial Intelligence, vol. 16, 2003, pp. 395-405.

[11] [12]

[13] [14]

[15] [16]

73

Md Zain, B. A., Tokhi, M. O., and Md Salleh, S., “PID control with genetic tuning of a single-link flexible manipulator”, The Sixteenth International Congress on Sound and Vibration, Krakow, 5-9 July 2009. S. Alavandar, T. Jain, and M.J. Nigam, “Hybrid bacterial foraging and particle swarm optimisation for fuzzy precompensated control of flexible manipulator”, International Journal of Automation and Control, vol. 4, no. 2, 2010, pp. 234-251. S. Mishra, “A hybrid least square-fuzzy bacterial foraging strategy for harmonic estimation”, IEEE Transactions on Evolutionary Computation, vol. 9 no. 1, 2005, pp. 61-73. T. Datta, T. et al., “Improved Adaptive Bacteria Foraging Algorithm in Optimization of Antenna Array for Faster Convergence”, Progress In Electromagnetics Research C, vol. l no. 1, 2008, pp. 143–157. Majhi, R. et al., “Efficient prediction of stock market indices using adaptive bacterial foraging optimization (ABFO) and BFO based techniques”, Expert Systems with Applications, vol. 36, 2009, pp. 10097-10104. Pandi, V. R. et al., “A hybrid bacterial foraging and differential evolution algorithm for congestion management”, European Transactions on Electrical Power 2009. H. Supriyono and M. O. Tokhi, “Bacterial Foraging Algorithm with Adaptable Chemotactic Step Size”, Second International Conference on Computational Intelligence, Communication Systems and Networks 2010, Liverpool-United Kingdom, 28-30 July 2010, pp. 72-77. DOI: 10.1109/CICSyN.2010.52 A. K. M. Azad, “Analysis and design of control mechanisms for flexible manipulator systems”, PhD thesis, Department of Automatic Control and Systems Engineering, The University of Sheffield, UK, 1994 G. G. Hastings and W. J. Book, “A linear dynamic model for flexible robotics manipulator”, IEEE Control Systems Magazine, vol. 7, 1987, pp.61-64.


Grid Workflow Recovery as Dynamic Constraint Satisfaction Problem Stanimir Dragiev, Joerg Schneider Department of Electrical Engineering and Computer Sciences Technische Universitaet Berlin Berlin, Germany {stanio,komm}@cs.tu-berlin.de

a valid schedule which meets the needs of the reservation, without breaking already accepted SLAs. As a next stage, a background scheduler optimises the effective schedule towards the policy of the administrative entity. The third stage is concerned with the handling of resource shortcomings resulting from failures. Dealing with failures has two aspects: support for jobs directly affected by a resource crash and support for jobs scheduled to start on the failed resource in the future. The first domain includes techniques like checkpointing, migration and restart, and enjoy much attention from researchers. However, our work is concerned with the second aspect of recovery scheduling, i.e., the remapping of all planned, but not yet started jobs affected by the outage. A fundamental problem in this domain is the uncertainty about the downtime of a given resource. In general, there are no reliable means to determine the nature of a failure at the time it happens; we can merely observe the result of it: the nonavailability of the resource. Under- or overestimation of the downtime leads to poor quality of the future schedules and thus to limited satisfaction of the administrative policy [2]. Hence, mechanisms are needed to adapt the reactions to the downtime. Recovery means, after all, to choose a valid schedule amongst large number of possible ones, whereby the choice is driven by feasibility criteria, job requirements and resource management policies. The domain concerned with precisely this kind of search problems is Constraint programming. Here, we propose to restate the recovery scheduling in terms of Constraint satisfaction problem (CSP) and benefit from approaches used in similar situations in other research areas. We are particularly interested in the concept of CSP with changing constraints over time, Dynamic CSP (DCSP), and the idea of maintaining similar consecutive solutions. Grid applications usually comprise several phases – data collection, processing, dissemination of results. These phases are realised by sub-jobs, which interact with each other. In other words, there are temporal and spatial dependencies between the single jobs, as well as QoS requirements on the resources responsible for running the jobs. The notion of Grid workflow [3] helps to consider all parts of a large application as a whole and to model the requirements in a

Abstract—With service level agreements (SLAs) the Grid broker guarantees to finish the Grid jobs by a given deadline. There are a number of approaches, to plan reservations to fulfil these deadline requirements and to handle currently running jobs in the case of a resource failure. However, there is a lack of strategies to handle the already planned but not yet started jobs. These jobs will be most likely also affected by the resource failure and can be remapped to other resources well in advance. Complex Grid jobs (Grid workflows) consisting of multiple sub-jobs introduce a higher complexity to determine a remapping saving as much Grid jobs as possible. In this paper a recovery scheme for Grid workflows using a dynamic constraint solver is presented and the gain in the number of saved Grid jobs is evaluated using extensive simulations.

I. I NTRODUCTION Grid computing opens lots of new perspectives for the application of the traditional information technologies. They range from on-demand high performance computational power for home usage to realisation of complex scientific applications involving multiple enterprises from different cultures, background, locations. At the same time, the new usage possibilities and the nature of the grid give rise to new challenges. We adopt a common view of the grid as a physically wide distributed collection of resources of different types which are spread across administrative and state boundaries. The idea to reproduce the organisational structure of the resources is the basis for the design [1] of the Virtual Resource Manager (VRM). It consists roughly of two layers: An administrative domain controller (ADC) is concerned with co-allocation and other grid-specific functions and allows for nested administrative domains (AD); Every managed resource is registered with the ADC via an active interface (AI) which addresses the heterogeneity by implementing the capabilities needed for participation in the grid and thus giving a unified view for all resources. The central question for the ADC is how to provide reliable information on the ability to process given reservation request, as well as to give QoS guarantees and to adhere to the negotiated service level agreements (SLA). To this end, the architecture of the ADC encompasses three stages of scheduling: Triggered by an incoming request for reservation, an online scheduler uses heuristics to produce

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

74

resource

Res R

B2

E1 A3

C2

C4 C3

Res Q

D1 A2

D4

D2

B3

B1

Res P

A1

E2 C1

t0

Terminate

D5

D3

E3 tremap

Cancel

May remap now

Need remap now

thorizon

No remap

time

Not considered

Figure 1. The remapping interval approach for failure recovery extended to support workflows and to limit the search space size. Workflows: A, B, C, D, E. Resources: P , Q and R; Resource P crashes in t0 . The approach extension cancels the jobs depending on terminated ones, and doesn’t consider the ones beyond thorizon .

consistent way. Important property of Grid workflows is that all sub-jobs contribute an essential part to the accomplishment of the final goal of the workflow, i.e. a workflow is successfully completed only if no jobs fail. In the following section, we comment known efforts to deal with scheduling and recovery in the grid, as well as CSP techniques for similar problems. After discussing our approach to determine a recovery schedule, we show the evaluation results.

consideration are concerned with directly affected jobs. The automatic recovery of future jobs affected by a failure is the topic of [12]. The authors propose an adaptive, load based downtime independent algorithm. The basic idea behind it is the introduction of a remapping interval which is calculated based on the current load situation. The resource is then assumed unavailable for this interval only and all jobs scheduled on the resource for this time are remapped. At the end of the interval the same procedure is repeated until the resource is up again. The remapping interval computation is based on the average incoming load per timeslot and currently booked load for every particular timeslot. Taken at a particular timeslot, their sum serves as estimation of the expected load in it. The remapping interval is determined as the timeslot for which the expected load falls below given level. This approach handles only independent jobs and is extended in this paper for Grid workflows. The well studied theory of Constraint satisfaction problems (CSP) can provide means for solving the problem of finding a valid schedule. However, a single CSP depicts the situation at a particular time. When resource availability changes, the constraints change, and this can be seen as new CSP instance. For handling such problems the notion of Dynamic constraint satisfaction problem was introduced

II. R ELATED W ORK The simple and robust approaches of managing resources, like providing only batch processing, are still widely used in practice, including MOAB [4], LSF [5], PBS [6], LoadLeveler [7], Globus Toolkit [8]. Despite their maturity, these techniques are not fully capable to ensure the QoS negotiated in service level agreements. The required advance is provided by middleware solutions incorporated in grid management systems like VRM [1] or by advanced local scheduler like CCS[9] or MAUI[10]. In [11] requirements for failure handling in the grid are defined. The emphasis is on the need of application context dependent recovery and distinct recovery on task and workflow level. The task level related techniques include checkpointing, migration, restart, and replication. All

75

Resource capacity

∀t∀r(

j X

st,r,j ≤ Rr .cap(t))

j=0

X X ∀j∀j 0 (wf l(j, j 0 ) → ( st,r,j = 0 → st,r,j 0 = 0))

Workflow atomicity

t,r

t,r

∀j∀r(∀tr ∀tf (rise(tr , j, r) < f all(tf , j, r))), rise(tx , r, j) := tx = −1 ∨ (0 ≤ tx < T − 1 ∧ stx ,r,j = 0 ∧ stx +1,r,j > 0), f all(tx , r, j) := tx = T − 1 ∨ (0 ≤ tx < T − 1 ∧ stx ,r,j > 0 ∧ stx +1,r,j = 0).

Connected assignment

Figure 2.

FOL representation of constraints. Examples.

(DCSP) [13]. The combination of Job shop scheduling problem and Minimal perturbation problem is the topic of [14]. It is based on the idea to maximize similarity between the broken and the new schedule. One way of maintaining minimal changes in solutions of consecutive CSPs is to produce schedules which are expected to remain valid after changes in the CSP [15]. For this approach to work properly it is assumed that certain types of expected changes are known which is limitedly applicable for environment with unexpected resource failures. An alternative idea is proposed in [16] – to take the previous, usually invalidated by constraint changes CSP solution as basis for the search of a new one. The algorithm works with two subsets of variables – the ones allowed to change and the fixed ones. The approach maximizes the similarity only with respect to the set of fixed ones. Several general purpose approaches which deal with overconstrained problems are discussed in [17], and this is, in general, the challenge in recovery scheduling. The guide introduces the notions of Fuzzy CSP, Probabilistic CSP, Weighted CSP Partial CSP (PCSP), constraint hierarchies and higher-order constraints (in other works referred to as reified constraints) . The latter allow for specifying predicates over constraints, like in ci or true. They seem to be the best supported ones in existing frameworks, for example, in the constrain solvers built upon Gecode [18] which we employ in our work.

leads to shorter intervals in heavier loads and longer intervals for more relaxed situations. For higher load, this forces the algorithm to eagerly search for possibilities to recover future jobs. Once the remapping interval is computed it determines the set of jobs we consider for remapping: the ones scheduled to start in the remapping interval – both on broken and intact resources. Figure 1 shows a synthetic schedule and the sorts of job from the recovery scheduler point of view. We consider the not completed jobs on the failed resource terminated. As consequence, their entire workflows are dropped from the schedule (workflows A and B in the example). The jobs scheduled to start in the remapping interval on the failed resource need to be relocated in order to be able to run (C1 , D3 ). By allowing a reordering of other jobs starting in the remapping interval (D2 , E1 , C2 ) we gain additional possibilities to obtain optimal schedule. B. Schedule Representation The considered jobs, their dependencies, the timeslots they run on, and the available resources define the space for the search for new schedule. This search space have to be represented in a form suitable for expressing constraints on it. We put it in terms of an integer problem: Definition 1: Let T := #{timeslots}, R := #{resources}, J := #{jobs}. Then, ST ×R×J is 3D integer matrix, whereas st,r,j = x ⇐⇒ ”in time slot t job j uses x of resource r”; with st,r,j ∈ {0, Jj .demand} In this representation, we can constrain the variables st,r,j and employ a constraint solver to find a valid solution.

III. A PPROACH The recovery scheduling is triggered by a resource failure which possibly makes the existing schedule infeasible. Roughly, the proposed algorithm does the following: determine a remapping interval, compute a horizon (the maximum considered time for the remapping of jobs), transform the current physical limitations and workflow requirements to constraints, pass the resulting CSP to a constraint solver, implement the found solution as new schedule.

C. Constraints The next task is to elaborate constraints which express the capacity limits of the local resources and the requirements of the accepted reservations. These are basically: Resource capacity: The sum of consumed entities of a resource cannot exceed the resource capacity. Enough resources in timeslot: In every particular timeslot in which a job runs, it has enough resources to run properly. Not split: Every job runs on one resource only. Workflow atomicity: A workflow depends on all its jobs, thus either all jobs can be scheduled, or no job at all. The latter removes needless load form the system.

A. Remapping interval The computation of the remapping interval is done according to [12]. The adaptation effect of the remapping interval is made dependent on the current load situation. Effectively, it

76

Temporal relations: If job jA is explicitly specified to run after a jB , the jobs should run in the given order. Parallel jobs: Jobs specified to be parallel need to have resources to run in the same timeslots as long as the longest job requires. Enough time: Jobs have to be assigned to a resource for enough timeslots. Types match: Jobs are assigned to resources capable of running them. Data dependencies: The data prerequisites of a job need to be satisfied, i.e. if job jB relies on input by job jA , they have to run in order jA and then jB , whereas if they do not run on the same resource, a transfer job JAB needs to be scheduled between them. Connected assignment: The timeslots a resource is assigned to a job have to be consecutive ones. Workflow times: All jobs start after the start time of the workflow and end before the workflow end time. The formal definition of the constraints is given in First order logic (FOL). As an example, consider the three constraints of figure 2. D. Over-constrained problem Obviously there are cases, a failure not only invalidates the current schedule but tightens the resource constraints so much that there is no schedule in which all accepted reservations fit. For such over-constrained problems, reification allows to relax selected constraints and to find the best available solution. Reification means to break a constraint, in order to make the problem solvable. This makes sense for constraints which do not represent physical limitations, i.e. only for the ones which determine whether a job runs at all. In our model, it is not feasible to reify, for example Resource capacity, or Workflow times. Both would lead to a schedule which is infeasible and thus to no benefit from the recovery. In contrast, the more jobs are dropped from the schedule, the easier to find valid assignment for the rest. In other words, if we allow to break constraints like Enough resources in timeslot we can adjust the load to the available resources. Effectively, this ensures the existence of solution and passes the responsibility for determining the optimal schedule to the objective function E. Optimisation Objective The notion of optimal schedule implies an objective for the search process. In order to counteract to the reification, we want as theP primary goal to maximize the scheduled load1 : W (S) = t,r,j st,r,j , with t spanning the indexes of the considered timeslots, r – the resource indexes, and j – the job indexes. As a secondary objective we impose on the solution the condition to remain as close as possible to the 1 this

Figure 3. Simulation results from the prototype implementation show the pre-failure load, the load after applying our approach and the load without recovery. Dotted lines show the confidence interval to the 99% level around the mean form multiple measurements.

may be seen as a measure for the earnings of the resource provider

77

P broken one: D(S, S 0 ) = t,r,j st,r,j − s0t,r,j . The latter is a similarity measure for schedules and constitutes the connection between consecutive CSP solutions and eventually utilises the quality of the old schedule.

phase, can lead to performance enhancements. Furthermore, formulation of the constraint in a portable way will allow for use of different constraint solvers and the strategies they implement.

IV. E VALUATION

R EFERENCES

To evaluate our approach, the recovery scheme was implemented as a module of the grid management system VRM [1]. We simulated a Grid with 8 computational resources, providing between 32 and 512 CPUs each. Sequences of randomly generated workflows were scheduled to simulate the normal operation. Then, one of the resources was assumed to fail, to start the recovery mechanism. We simulated a variety of parameters which we expected to impact the performance: load, job length, time available for recovery, number of jobs per workflow. The experiments in Figure 3 already show the significant advance of the recovery mechanism compared to naively dropping all affected future jobs and the workflows they belong to. In addition, the approach is often able to preserve more than 80% of the pre-downtime load. All pictures show on the y-axis the preserved load normalized to the total load before the resource failure, with respect to varying parameters on the x-axis. The upper line is the total pre-downtime load itself – constantly one. The lower line is the load remaining in the schedule after the failure without any recovery mechanism. The middle line represents the proposed recovery scheme using DCSP. The four experiments depicted in Figure 3 show that the performance of the recovery scheme is independent of the system parameter. By changing the interarrival time we modelled highly utilized grids (low interarrival time) and lower utilized Grids. In the upper right diagram we modified the complexity of the Grid workflows to be scheduled, which obviously has an impact at the loss if no recovery is done but hardly any impact on the recovery scheme. In the lower row we modified the size of the jobs in the resource and time dimension.

[1] L.-O. Burchard, M. Hovestadt, O. Kao, A. Keller, and B. Linnert, “The virtual resource manager: An architecture for SLAaware resource management,” in 4th Intl. IEEE/ACM Intl. Symposium on Cluster Computing and the Grid (CCGrid) 2004, Chicago, USA, 2004, pp. 126 – 133. [2] Burchard, L.-O. and B. Linnert, “Failure Recovery in Distributed Environments with Advance Reservation Management Systems,” in 15th IFIP/IEEE International Workshop on Distributed Systems: Operations and Management (DSOM), Davis, USA, ser. Lecture Notes in Computer Science (LNCS), vol. 3278. Springer, 2004, pp. 112–123. [3] D. Hollingsworth, “The workflow reference model version 1.1,” Workflow Management Coalition, Tech. Rep. WFMCTC-1003, January 19th 1995. [4] MOAB Team, “MOAB workload manager,” on-line, 2009, http://www.clusterresources.com/products/moab-cluster-suite. [5] P. C. Inc., “Platform LSF,” on-line, http://www.platform.com/products/LSFfamily.

2009,

[6] Altair Grid Technologies, “PBS Pro administrator guide 5.4,” on-line, 2004. [7] IBM, “Tivoli workload scheduler LoadLeveler,” on-line, 2009, http://ibm.com/systems/clusters/software/loadleveler. [8] I. Foster and C. Kesselman, “Globus: A metacomputing infrastructure toolkit,” International Journal of High Performance Computing Applications, vol. 11, no. 2, p. 115, 1997. [9] CCS Team, “CCS: Computing Center Software,” on-line, 2009, https://www.openccs.eu/core/. [10] R. K. Brett Bode, David M. Halstead and Z. Lei, “The portable batch scheduler and the maui scheduler on linux clusters,” in Proceedings of the 4th Annual Linux Showcase & Conference. USENIX Association, 2000.

V. S UMMARY AND OUTLOOK We presented an approach to recover Grid workflows after resource failure. It focuses on the future jobs scheduled on the failed resource. Furthermore, we propose to employ a constraint solver for the search for optimal schedule. To this end, a formal representation of a schedule as an integer matrix is elaborated, as well as a constraint system which reflects the feasibility requirements for a schedule. The objective function for the optimisation includes maintaining similarity to previous solutions. Future research will go in multiple directions: An alternative schedule representation can bring rigorous improvements in the complexity of constraint description and in the computational complexity. Study of the internal parameters of the approach, including the use of heuristics in the search

[11] S. Hwang and C. Kesselman, “Grid workflow: A flexible failure handling framework for the grid,” Proceedings of the 12th IEEE International Symposium on High Performance Distributed Computing (HPDC03), 2003. [12] L.-O. Burchard, B. Linnert, and J. Schneider, “A distributed load-based failure recovery mechanism for advance reservation environments.” in Cluster Computing and the Grid, 2005. CCGrid 2005. IEEE International Symposium on, vol. 2, May 2005, pp. 1071–1078. [13] R. Dechter and A. Dechter, “Belief maintenance in dynamic constraint networks,” in Proceedings of the 7th National Conference on Artificial Intelligence, R. G. Smith and T. M. Mitchell, Eds. St. Paul, Minnesota: Morgan Kaufmann, Aug. 1988, pp. 37–42.

78

[14] Y. Ran, N. Roos, and J. van den Herik, “Methods for repair based scheduling,” 2002, proceedings of the 21th workshop of the UK PLANNING and SCHEDULING Special Interest Group,PLANSIG 2002 (2002) 79-86.

[16] G. Verfaillie and T. Schiex, “Solution reuse in dynamic constraint satisfaction problems,” in Proceedings of the National Conference on Artificial Intelligence (AAAI 94), 1994, pp. 307–312.

[15] R. J. Wallace and E. C. Freuder, “Stable solutions for dynamic constraint satisfaction problems,” in Principles and Practice of Constraint Programming - CP98, 4th International Conference, Pisa, Italy, October 26-30, 1998, Proceedings, ser. Lecture Notes in Computer Science, M. J. Maher and J.-F. Puget, Eds., vol. 1520. Springer, 1998, pp. 447–461.

[17] R. Bartak, “On-line guide to constraint programming, 1st edition; 1st part: Constraint satisfaction,” on-line, 2009, http://ktiml.mff.cuni.cz/ bartak/constraints. [18] Gecode Team, “Gecode: Generic constraint development environment,” on-line, 2009, http://www.gecode.org.

79


Speckle Filtering Of Ultrasound B-Scan Images A Comparative Study Between Spatial And Diffusion Filters R. Sivakumar1, M. K. Gayathri2 and D. Nedumaran3 1, 2 & 3

Central Instrumentation & Service Laboratory, University of Madras, Chennai, India E-mail: [email protected], [email protected]

Abstract: Speckle noise is the inherent property of ultrasound B-Scan images which has been filtered using well-established speckle reduction techniques. In this work, six spatial filters namely Frost, Median, Lee, Kuan, Wiener, and Homomorphic filters, and two diffusion filters viz., Speckle Reduction Anisotropic Diffusion (SRAD) filter, and Anisotropic Diffusion (AD) filter have been attempted over 200 different digital ultrasound B-scan images of kidney, abdomen, liver and choroids. A comparative study has been made on these filters in preserving the edges of the images with effective denoising by calculating fourteen established performance metrics along with the execution time in order to determine the effective and optimum despeckling algorithm for real time implementation. To do this, a cumulative speckle reduction (CSR) algorithm has been developed using MATLAB 7.1, which performs all despeckle filtering functions as well as performance metrics calculation in a single iteration. This study reveals that most of the despeckle filters performed well and gave optimum performance, but SRAD is the outperformed filtering technique for B-scan ultrasound image as far as the performance metrics, execution time and visual inspection are concerned. Key words: B-Scan image, Speckle, Spatial Adaptive Filters, Anisotropic Diffusion Filters, SRAD, Matlab, Performance metrics. I. INTRODUCTION

Medical ultrasound B-scan imaging has been used for effective diagnostics of diseases over the past decades due to its noninvasive, harmless, portable, accurate and cost-effective characteristics [1-5]. Unfortunately, the quality of ultrasound B-scan image is generally limited by the noise called speckle. Speckle noise occurs when a coherent source and a noncoherent detector are used to grill a medium, whose surface is rough on the scale of a typical ultrasound wavelength. Especially, speckle noise occurs in the images of soft organs such as liver and kidney whose underlying structures are too small to be resolved by the large ultrasound wavelength. Several papers have been found in the literature describing the properties [6-8], modeling, and analysis and filtering [9-14] of speckle noise in ultrasound images. In this paper, ultrasound B-scan image will be analyzed as coherent speckle using the mathematical model [15] given in equation. 1. R( x, y ) = N ( x, y )η M ( x, y ) + η A ( x, y )

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

(1)

80

where R( x, y) is the real noisy image, N ( x, y) represents an unknown noise free image, η M ( x, y) and η A ( x, y ) are multiplicative and additive noise respectively. II. SPECKLE FILTERING TECHNIQUES

Over the years, several techniques have been developed to despeckle images with speckle noise [16]. Speckle filtering consists of a kernel moving over each pixel in the image and calculating and replacing the central pixel values under the kernel. The kernel is moved along the image one pixel at a time until the entire image has been covered. In this work, spatial adaptive filtering and anisotropic diffusion filtering techniques have been employed to study the speckle noise reduction in ultrasound B-scan images. A. SPATIAL ADAPTIVE FILTERING

Spatial Adaptive filter takes a moving filter window and calculates the statistical information of all pixels gray values such as the local mean and the local variance. The central pixel’s output value is dependent on the calculated statistical information. Spatial filters like Lee-filter, sigma-filter, and Kuan-filter were the earliest filters working directly on the intensity of the image using local statistics [12-14]. Also, computation of local statistics, region growing procedure and application of smoothening operator are the three main steps involved in the implementation of spatial adaptive speckle filtering [9]. This section describes the brief definition and mathematical description of the spatial adaptive filters. 1. Frost filter: The Frost filter is an adaptive and exponentially weighted averaging filter based on the coefficient of variation, which is the ratio of the local standard deviation to the local mean of the degraded image [17]. It replaces the pixel of interest with a weighted sum of the values within the moving kernel and the weighting factors decrease with distance and increase in accordance with the increase in variance of the kernel. This filter assumes speckle as multiplicative noise and stationary noise, which follows the statistics given in equation 2. DN = ∑ Aδ e n×n

−α t

(2)



 2 

where δ =  4   σ   nσ ′2   I ′2 

(3)

pass or the high-boost Butterworth filter. In this study, the high-boost Butterworth technique has been employed and is given in equation 7.

where A is the normalization constant, I`2 is the local mean, σ is the local variance and σ`2 is the image coefficient of variation value, t = X − X 0 + Y − Y0 and n is the moving kernel size.

H u ,v = δ L +

δH  D 1+  0  Du ,v 

   

2

Du,v = (u − N / 2)2 2 + (v − N / 2)2

2. Median filter: The median filter [17] is a spatial non-linear filter which removes pulse or spike noise by replacing the middle pixel value in the window with the median value of its neighbors in the window. 3. Lee and Kuan filter: The Lee [12, 13] and Kuan [14] filter are based on the minimum mean square error (MMSE), which produce the speckle free image governed by the relationship given in equation 4 [18]. U ( x, y ) = I ( x, y )W ( x, y ) + I ′( x, y )(1 − W ( x, y ))

(4)

where I ′ is the mean value of the intensity within the filter window, and W(x,y) is the adaptive filter coefficient calculated using the following formula. 2 CB 2 CI2 + CB C2 1− B CI2 W ( x, y ) = 2 1 + CB

W ( x, y ) = 1 −

for Lee filter

(5)

for Kuan filter

(6)

where CI is the coefficient of variation of the noised image and CB is the coefficient of variation of the noise. In general, the value of W(x,y) approaches zero in uniform areas, i.e., it approaches unity at edges which results in little modification of pixel values near edges. 4. Wiener filter: Wiener filtering is a method [15, 19] of restoring images in the presence of blur as well as noise. Wiener filter performs smoothing of the image based on the computation of local image variance. When the local variance of the image is large, the smoothing is little. On the other hand, if the variance is small, the smoothing will be better. This approach often produces better quality results than linear filtering, since the Wiener filtering is adaptive, more selective than a comparable linear filter. It preserves edges and other high-frequency information of the image, but requires more computation time than linear filtering. 5. Homomorphic Filtering: Homomorphic filtering performs image enhancement by applying the filter function and inverse FFT on the logarithmic compressed image [20]. The filter function H(u,v), may be constructed using either the band-

81

(7)

(8)

where D0 is the cut of frequency of the filter, δL is the low frequency gain, δH is the high frequency gain, u and v are the spatial coordinates of the frequency transformed image and N is the dimensions of the image in the u and v space. B. DIFFUSION FILTER

Diffusion filters may be applied directly on the image for removing the speckle noise by solving partial differential equation. An Anisotropic diffusion performs contrast enhancement and noise reduction without requiring the power spectrum information of the image [20]. In this work, two diffusion filters have been attempted and their descriptions are given here. 1. SRAD filter: Speckle Reducing Anisotropic Diffusion (SRAD) [21] filter eliminates speckle without distorting and destroying useful image information and the important image edges [22] respectively. The SRAD exploits the instantaneous coefficient variation in reducing the speckle. 2. Anistropic Diffusion filter: In this study, Perona and Malik [23-25] Anistropic Diffusion (PMAD) method based on the nonlinear partial differential equation (PDE) is used and its mathematical functions are given in equation 9 and 10. ∂I = div c ( ∇I ) ⋅ ∇I  ∂t

(9)

I (t = 0) = I 0

(10)

where ∇ is the gradient operator, the ‘div’ divergence denotes the magnitude, c(x) is the diffusion operator, coefficient, and I0 is the initial image. They suggested two diffusion coefficients and are given in equations 11 and 12. 1 2 x 1+   k   x 2  c( x) = exp  −     k    c( x) =

and

(11)

(12)

where k is an edge magnitude parameter. In this method, the image edge or boundary can be detected using gradient magnitude. The discrete form of the equation (9) is given by

(

)

∆t t t I st +∆t = I st + ∑ c ∇I s, p ∇I s, p ηs p∈η s

III. PARAMETERS/METRICS FOR ANALYZING DESPECKLE FILTER PERFORMANCE

(13)

where I st is the sampled image, s is the pixel position in a discrete two-dimensional (2-D) grid, and ∆t is the time step size, η s represents the spatial neighborhood of pixel s, η s is the number of pixels in the window, and ∇I st , p = I tp − I st , ∀p ∈η s .

Performance Metrics Average Difference (AD)

Mean Square Error (MSE)

Peak Signal to Noise Ratio (PSNR) Maximum Difference (MD)

To quantify the performance of the despeckle filter algorithm in terms of the efficiency of removing the speckle noise and enhancing the useful image information, the following established performance metrics found in the literatures [2635] are calculated in this study and their mathematical expression, definition, significance range are tabulated in Table I. All the metrics are self explanatory and hence a separate explanation for each and every metrics is not included in the discussion due to page limitation.

TABLE I PARAMETERS CALCULATED FOR DESPECKLING FILTERS INDICATING THE QUALITY OF THE IMAGE Range of Value for better Performance Mathematical Expression Definition (Min/Max/ Close to Unity) AD is maximum for dissimilar images Mean difference of the two 1 M N ' AD = ∑ ∑ ( X j , k − X j, k ) and minimum for similar images. The (Original and denoised) images MN j =1 k =1 range of AD is from 0 to 255. divided by the size of the image Higher and lower MSE values indicate The average of the square of the larger and smaller differences between difference between original and the original and denoised image, M N 1 ' 2 the denoised image divided by the MSE = ∑ ∑ ( X j , k − X j ,k ) respectively. MSE will be equal to zero MN j =1 k =1 size of the image (the error). for identical images. For completely Represents the average difference dissimilar images, the MSE value between images. becomes 255. Measure of the performance of the Typical value is between 30 and 50 dB. speckle noise removal. It is a ratio Higher PSNR values show better image  2552  (2n − 1) n  PSNR = 10 log10 = 10 log10  quality. For identical images, the MSE between the maximum possible  MSE  MSE become zero and the PSNR is power of the signal and the noise   undefined. content. Maximum error difference MD gives the maximum difference MD = Max( X j ,k − X 'j ,k ) between the original and denoised values in the pixel level image

Normalized Absolute Error (NAE)

M N ' ∑ ∑ X j , k − X j ,k j =1 K =1 NAE = M N ∑ ∑ X j ,k j =1 K =1

Normalized absolute error is the measures of error prediction accuracy of the image

Its value ranges between 0 and 1. Lower value indicates that the error between the original and denoised image is smaller.

Structural Content (SC)

M N 2 ∑ ∑ X j, k j =1 K =1 SC = M N '2 ∑ ∑ X j ,k j =1 K =1

Measure of similarity between original and denoised image.

For identical images it should be one.

Measure of edge preservation in the denoised image.

Its value is 1 and 0 for identical and completely uncorrelated images, respectively. Its value becomes –1 if they are completely anti-correlated.

Coefficient of Correlation (CoC)

Normalized Cross Correlation (NCC)

Image Quality Index (IQI)

Speckle Index (SI) Average Signal to Noise Ratio (ASNR)

CoC =

M N ' ' ∑ ∑ ( X j , k −X j , k )( X j , k − X j , k ) j =1 K =1 M N ' ' 2M N 2 ∑ ∑ ( X j,k −X j,k ) ∑ ∑ ( X j,k − X j,k ) j =1 K =1 j =1 K =1 M N ' ∑ ∑ ( X j ,k )( X j ,k ) j =1 K =1 NCC = M N 2 ∑ ∑ X j,k j =1 K =1

IQI =

It is a correlation based image quality measure.

XX ' xx'   σ x2 + σ 2   X 2 + ( X ' )2  x'       4σ

M N 1 σ ( j, k ) SI = ∑ ∑ MN j =1 k =1 µ ( j , k ) ASNR =

1 Speckle Index

82

Its value becomes unity for identical images.

Degree of distortion in terms of loss of correlation, mean distortion and variance distortion.

Its dynamic range is between -1 and 1. For identical images the value of image quality index is unity.

Measure of speckle removal in terms of average contrast of the image

If SI is low then the image quality is improved.

Measures average deviation of the speckle with respect to the mean value of the image.

Less for noisy images and increases with the degree of denoising.

Image Variance (IV) Noise Standard Deviation (NSD)

IV =

1 M −1 N −1 )2 −µ ∑ ∑ (P j,k MN j =0 k =0 j, k

NSD =

Effective Number of Looks (ENL) Mean Structure Similarity Index Map (MSSIM) Structure Similarity Index Map (SSIM)

M N ' 2 ∑ ∑ ( X j , k − NMV ) j =1 k =1 MN ENL =

MSSIM =

[ NMV ]2 [ NSD]2

1 M N ' ∑ ∑ SSIM [(X j , k ), ( X j , k )] MN j =1 K =1

where SSIM ( X , X ' ) =

(2 µ X µ ' + C1)(2σ + C2 ) X XX ' ( µ 2X + µ 2 + C1)(σ 2X + σ 2 + C2 ) X' X'

Determines the contents of the speckle in the image.

A lower value gives smoother image as more speckles is reduced.

Determines the quantity of the speckle in the image

NSD value will be less for the images with minimum quantity of the speckle.

Measure of speckle level in ultrasound image over a uniform image region.

A large value of ENL reflects the better quantitative performance of the filter. The value also depends on the size of the testing region.

MSSIM and SSIM are used to compare luminance, contrast and structure of two different images. It can be treated as a similarity measure of two different images

The MSSIM value should be closer to unity for optimal measure of similarity.

IV. IMPLEMENTATION OF DESPECKLE ALGORITHM AND RESULTS

comparable with the outperformed SRAD filter. But SRAD has exceptional performance than Wiener in terms of metrics like AD, MD, MSE, NAE, IV and visual inspection.

In this study, the cumulative speckle reduction (CSR) algorithm comprising of six traditional spatial filters and two diffusion filters with fourteen qualitative metrics estimation, has been developed in the MATLAB environment. More than 200 digital ultrasound B-scan images of organs like kidney, choroids, abdomen and liver were obtained for several cases from the GE healthcare ultrasound machine (Model: VIVID7). The CSR algorithm was tested in all the 200 digital B-scan images. In all these trails, the CSR algorithm provided the despeckled images of all the eight filters and fourteen metrics of the respective filter with execution time of the algorithm in a single iteration successfully. Of the 200 tested images, the original and despeckled B-scan images and their SSIM images of the choroids obtained in the test run are given in fig. 1 and fig. 2, respectively. Further, the various performance metrics calculated for the despeckled image of the choroids are given in Table II. From the Fig.1, 2 and Table II, some of the important interpretations arrived in this study are given here: 1.

Frost and Median filters slightly improve the information of the edges, but the Lee filter improves the ability of preserving the edges.

2.

Kuan and Frost filters despeckle the image to some extent. But it can also be found from the metrics that Frost filter is better than the Kuan filter for speckle reduction since it has higher SNR, SC, and ENL values than Kuan filter.

3.

Homomorphic filter sharpens the image and flattens the speckle variation since it has very low PSNR and high MD values.

4.

Wiener is the better approach as per as spatial filters are concerned, since it has performance metrics

5.

PMAD filter in B-scan images enhances the speckle noise rather than suppressing it. The parameters like PSNR, SC, NCC, IV clearly depicts that the PMAD filter is not an efficient filter for the ultrasound images. Also, the image quality index Q and SSIM decreases as the iteration value increases which shows that the image is distorted rather than enhanced to a great extent.

6.

The SRAD suppresses the speckle noise to a great extent interms of higher PSNR, larger ENL, lower NAE, lower SI, and lower NSD values found from Table II. The performance metrics AD, MSE, MD, NAE, SC, ASNR, NSD increases as the iteration value increases. The values of PSNR, CoC, NCC, Q, SI, IV, ENL, MSSIM decreases as the iteration increases. As the iteration of the SRAD is greater than 8, the image gets blurred. Also, SRAD doesn’t give noticeable difference in despeckling in comparison with other filters, when the iteration is less than 4. As a result, the optimum values of iteration found in this study for better despeckling is in the range of 4-8.

Thus, all the spatial and diffusion filter performance were compared and tested experimentally for various digital ultrasound B-scan images interms of performance metrics. From the performance metrics and visual inspection of the zoomed/enlarged interested regions of the despeckled images, it has been verified that the SRAD filtering approach is performing better than the traditional spatial filters. As a result, SRAD exhibits optimal filtering performance for speckle reduction in B-scan ultrasound images.

83

for feature extraction and excels over the traditional despeckle filters and the conventional anisotropic diffusion method in terms of speckle reduction, edge preservation and image clarity. (a)

(d)

(b)

(e)

(c)

ACKNOWLEDGMENT

We acknowledge the help extended by Prof. Dr. S. Thanikachalam, Chairman, Cardiac Care Centre, SRMC Hospital, Chennai for providing various ultrasound B-scan images to carry out this study. Also, we record the financial assistance provided by Tamilnadu State Council for Science Technology (TNSCST), Chennai to perform this work.

(f)

REFERENCES (i) (g) (h) Fig. 1 View of the original and despeckled choroids images (a) Original (b) Frost filter (c) Median filter (d) Lee filter (e) Kuan filter (f) Wiener filter (g) Homomorphic filter (h) SRAD filter (i) PMAD filter

[1] [2] [3]

[4]

(a)

(b)

[5]

(c)

[6]

[7] (d)

(e)

(f) [8] [9]

(g)

(h)

(i) [10]

Fig. 2 View of the original and SSIM factor images of the choroids (a) Original (b) Frost filter (c) Median filter (d) Lee filter (e) Kuan filter (f) Wiener filter (g) Homomorphic filter (h) SRAD filter (i) PMAD filter V. CONCLUSIONS

[11]

In this paper, we have developed a cumulative speckle reduction (CSR) algorithm for six spatial and two diffusion type filters with fourteen performance metrics and execution time for the estimation of the optimum despeckle filter for real-time application. The algorithm was developed in MATLAB 7.1 and tested in more than 200 ultrasound Bscan images of four different organs viz liver, abdomen, choroids and kidney. In all these images, the algorithm performs well and produces performance measuring parameters depending on the image content within the limited range consistently. Further, the algorithm generates all the eight filter outputs as well as fourteen performance metrics in a single iteration. Based on the results obtained and the visual inspection of the despeckled ultrasound Bscan image outputs, we conclude that the diffusion filters exhibits a highly efficient noise reduction and the ability to preserve and even enhance the edges of the images in comparison with the spatial filters. Moreover, the SRAD filter is shown to generate B-scan images with better quality

[12]

[13]

[14]

[15] [16]

[17]

[18]

84

Streve Webb, “The Physics of Medical imaging, Medical Science Series,” Taylor and Francis Group, pp.319-377, 1988. Kayvan Najarian and Robert Splinter, “Biomedical Signal and image processing,” Taylor and Francis publications, pp.333-364, 2006. Chris Guy and Dominic ffytche, “An Introduction to the principles of medical imaging. Revised edition,” Imperial College Press, pp.267-294, 2005. William R. Hendee and E. Russell Ritenour, “Medical Imaging Physics,” fourth edition, John Wiley and Sons, Inc., Publication, pp.317338, 2002. Thomas L. Szabo, “Diagnostic Ultrasound imaging, Academic Press Series in Biomedical Engineering,” Elsevier Academic Press, 2004. Robert F. Wagner, Stephen W. Smith, John M. Sandrik, and Hector Lopez, “Statistics of Speckle in Ultrasound B- Scans,” IEEE Trans. on Sonics and Ultrasonics, Vol. 30, No. 3, pp.156-163, May 1983. J.W. Goodman, “Some Fundamental Properties of Speckle,” Journal of Optical Society of America, Vol.66, No.11, pp.1145-1150, November 1976. Christoph B. Burckhardt, “Speckle in Ultrasound B-Mode Scans,” IEEE Trans. on Sonics and Ultrasonics, Vol.25, No.1, pp1-6, 1978. Mustafa Karaman, M. Alper Kutay, and Gozde Bozdagi, “An Adaptive Speckle Suppression Filter for Medical Ultrasonic Imaging,” IEEE Trans. On Medical Imaging, Vol. 14, No.2, pp.283-292, 1995. Victor S. Frost, Josephine Abbott stiles, K. S. Shanmugan, and Julian C. Holtzman, “A Model for Radar Images and Its Application to Adaptive Digital Filtering of Multiplicative Noise,” IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. PAMI-4, No.2, pp.157-166, March 1982. Alin Achim, Anastasios Bezerianos and Panagiotis Tsakalides, “Novel Bayesian Multiscale Method for Speckle Removal in Medical Ultrasound Images,” IEEE Trans. on Medical Imaging, Vol. 20, No. 8, pp.772-783, August 2001. J. S. Lee, “Digital Image Enhancement and Noise Filtering by Use of Local Statistics,” IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol. PAMI-2, No. 2, pp.165-168, March 1980. J. S. Lee, “Refined filtering of image noise using local statistics, Computer Vision,” Graphics, and Image Processing. Vol.15, pp.380389, 1981. D. T. Kuan, A. A. Sawchuk, T.C. Strand and P. Chavel, “Adaptive restoration of images with speckle,” IEEE Trans. on Acoustics, Speech and Signal Processing, Vol. ASSP-35, No.3, pp.373-383, March1987. Anil K. Jain, “Fundamentals of Digital Image Processing,” first edition, Prentice-Hall, Inc, 1989. Stian Solbo and Torbjorn Eltoft, “Homomorphic Wavelet-Based Statistical Despeckling of SAR Images”, IEEE Trans. on Geoscience and Remote Sensing, Vol. 42, No. 4, pp. 711-721, April 2004. Raman Maini, and Himanshu Aggarwal, “Performance Evaluation of Various Speckle Noise Reduction Filters on Medical Images,” International Journal of Recent Trends in Engineering, Vol.2, No. 4, pp.22-25, November 2009. Nadia Souag, “Speckle Reduction in Echocardiographic Images,” 14th European signal processing conference (EUSIPCO 2006) Proceeding, Florence, Italy, Sept 4-8, 2006.

[19] Rafael C. Gonzalez and Richard E. Woods, “Digital Image Processing,” Third Edition, Pearson Education, 2008. [20] Christos P. Loizou, Constantinos S. Pattichis, Christodoulos I. Christodoulou, Robert S. H. Istepanian, Marios Pantziaris, and Andrew Nicolaides, “Comparative Evaluation of Despeckle Filtering In Ultrasound Imaging of the Carotid Artery,” IEEE Trans. on Ultrasonics, Ferroelectrics, and Frequency control, Vol. 52, No. 10, pp.1653-1669, October 2005. [21] Byeongcheol Yoo and Toshihiro Nishimura, “A Study of Ultrasound Images Enhancement using Adaptive Speckle Reducing Anisotropic Diffusion,” IEEE International Symposium on Industrial Electronics (ISlE 2009) July 5-8, Seoul Olympic Parktel, Seoul, Korea, pp.581-585, 2009. [22] Yongjian Yu and Scott T. Acton, “Speckle Reducing Anisotropic Diffusion,” IEEE Trans. on Image Processing, Vol. 11, No. 11, pp.12601270, November 2002. [23] P. Perona and J. Malik, “Scale space and edge detection using anisotropic diffusion,” IEEE Trans. on Pattern Analysis and Machine Intelligence, Vol.12, No.7, pp. 629–639, JULY 1990. [24] Jesse S. Jin, Yung Wang, and John Hiller, “An Adaptive Nonlinear Diffusion Algorithm for Filtering Medical Images,” IEEE Trans. on Information Technology in Biomedicine, Vol. 4, No. 4, pp. 298-305, December 2000. [25] Khaled Z. Abd-Elmoniem, Abou-Bakr M. Youssef, and Yasser M. Kadah, “Real-Time Speckle Reduction and Coherence Enhancement in Ultrasound imaging via nonlinear anisotropic diffusion,” IEEE Trans. on Biomedical Engineering, Vol. 49, No 9, pp.997-1014, September 2002.

[26] C. I. Christodoulou, C. Loizou, C. S. Pattichis, M. Pantziaris, E. Kyriakou, M. S. Pattichis, C. N. Schizas and A. Nicolaides, “Despeckle filtering in ultrasound imaging of the carotid artery,” Proceedings of the second joint EMBS/BMES Conference, Texas, USA, pp.1027-1028, October 23-26, 2002. [27] Sonja Grgic, Mislav Grgic and Marta Mrak, “Reliability of objective picture quality measures,” Journal of Electrical Engineering, Vol.55, No.1-2, pp.3-10, 2004. [28] Ahmet M. Eskicioglu and Paul S. Fisher, “Image quality measures and their performance,” IEEE Trans. on Communications, Vol.43, No.12, pp.2959-2965, Dec.1995. [29] Zhou Wang and Alan C. Bovik, “A Universal image quality index,” IEEE Signal Processing Letters, Vol.9, pp.81-84, March 2002. [30] Rajeev Srivastava J.R.P Gupta and Harish Parthasarthy, “Comparison of PDE based and other techniques for speckle reduction from digitally reconstructed holographic images,” Elsevier, Journal of Optics and Lasers in Engineering, Vol.48, pp.626-635, 2010. [31] Karunesh K. Gupta and Rajiv Gupta, “Despeckle and geographical feature extraction in SAR images by wavelet transform,” Elsevier, ISPRS Journal of Photogrammetry and remote sensing, Vol.62, pp.473484, July 2007. [32] Zhou Wang, Alan Conrad Bovik, Hamid Rahim Sheikh and Eero P. Simoncelli, “Image Quality assessment: from Error visibility to structural similarity,” IEEE Trans. on Image Processing, Vol.13, No.4, pp.600-612, April 2004. [33] Ismail Avcibas, Bulent Sankur and Khalid Sayood, “Statistical evaluation of Image quality measures,” Journal of Electronic Imaging, Vol.11, No.2, pp.206-223, April 2002. TABLE II CALCULATED PERFORMANCE METRICS OF THE VARIOUS DESPECKLING FILTERS IMPLEMENTED IN THE CSR ALGORITHM FOR CHOROIDS B-SCAN IMAGE Note: (O) & (F) denote original and denoised images respectively

Performance Metrics AD

Name of the Filter Frost

Median

Lee

Kuan

Wiener

Homomorphic

SRAD

PMAD

0.108443

0.26998

0.180632

0.197123

0.013133

67.343803

0.002167

6.998079

MSE

42.359813

46.67553

82.36125

83.37816

22.76085

7029.658956

16.136394

100.325448

PSNR

31.861263

31.43991

28.40253

28.92028

34.558917

9.661461

36.052739

28.116693

MD

136

241

162

162

29

253.17971

43

54

NAE

0.052439

0.045748

0.064168

0.065182

0.047998

0.977485

0.03814

0.101575

SC

1.011429

1.018497

1.024863

1.025055

1.009394

2990.22

1.01303

0.85073

CoC

0.991581

0.99071

0.978536

0.983335

0.995488

0.942364

0.996901

0.991464

NCC

0.99143

0.987702

0.981123

0.982031

0.993778

0.01561

0.992457

1.080815

Q

0.876071

0.871302

0.82985

0.83863

0.863715

0.000541

0.942106

0.857614

SI(O)

3.26E-6

3.26E-6

3.26E-6

3.26E-6

3.26E-6

3.26E-6

3.26E-6

3.26E-6

SI(F)

3.05E-6

3.17E-6

3.08E-6

3.08E-6

3.21E-6

3.19E-8

3.04E-6

3.19E-6

ASNR (O)

3.06E+5

3.06E+5

3.06E+5

3.06E+5

3.06E+5

3.06E+5

3.06E+5

3.06E+5

ASNR (F)

3.17E+5

3.14E+5

3.2E+5

3.2E+5

3.1E+5

3.13E+7

3.29E+5

3.13E+5

IV(O)

0.932

0.932

0.932

0.932

0.932

0.932

0.932

0.932

IV(F)

0.927

0.926

0.947

0.943

0.928

1.117

0.917

0.923

NSD(O)

49.173

49.173

49.173

49.173

49.173

49.173

49.173

49.173

NSD(F)

44.013

46.681

44.943

44.921

41.224

61.019

38.131

40.327

ENL(O)

1.316

1.316

1.316

1.316

1.316

1.316

1.316

1.316

ENL(F)

1.549

1.394

1.494

1.486

1.617

0.938

1.784

1.652

MSSIM

0.910605

0.905621

0.878193

0.885086

0.909313

0.052414

0.958184

0.896904

Execution time

11.376117

0.653691

15.011232

14.777647

0.43834

0.4366

0.628065

0.643238

85


Development of Dashboard for Hospital Logistics Management 1

2

Mahendrawathi ER , Danu Pranantha , Johansyah Dwi Utomo Information System Department, Faculty of Information Technology, Institut Teknologi Sepuluh Nopember Kampus ITS, Keputih, Sukolilo 60111 Surabaya, Indonesia [email protected], [email protected] performance. Based on this information they can then identify area for improvements. Logistics aspects have not gained enough attention in hospital management to date. Research in hospital management mainly addresses the improvement of hospital services through simulation [3, 4, 5, 6, 7, 8] and scheduling of doctors and nurses [9, 10, 11]. Tung et al [1] is among a few work that address hospital logistics. However, it only focuses on investigating the use of information systems in managing hospital logistics. It does not specifically deal with the management of logistics in a hospital.

Abstract- Logistics as an important process in providing high quality and responsive health services has not received enough attention in the literature to date. This paper presents an initial work that attempt to fill this gap by developing a prototype of a dashboard for logistics management in hospital XYZ. The methodology used in this research is divided into intelligence, design and implementation phases. The intelligence phase starts with literature study, survey, gathering objectives and Key Performance Indicator, and collecting necessary data. The design phase consists of designing and constructing database, defining presentation model for each KPI, and designing the storyboard. The dashboard is developed based on the design in the implementation phase. Implementation using several items as samples has demonstrated the potential use of the dashboard. It was found that the item category that contributes highly to the total inventory value (50%) of the hospital has a low service level. Further investigation shows that one item in this category has very low service level because the amount of item received is lower than demand. These findings can be used by management as starting point to identify the root cause and take the necessary actions to solve the problem.

I.

This paper presents the work that attempt to fill part of the gap by developing a prototype of a dashboard for logistics management in hospital XYZ. The dashboard is designed to display relevant information on the service level of key items in an informative and interesting single display. The dashboard will enable hospital manager to evaluate service level of key items and quickly identify those that do not meet the targets. This can be used as a basis to identify problems and solutions in logistics management.

INTRODUCTION

II. DASHBOARD

A hospital is required to provide the highest service to its patient in affordable cost. One of the processes in hospital management that plays a crucial role in providing high quality and responsive health services is logistic process. Logistics management includes materials management and fulfillment, instrument supply and management, and procurement [1]. Logistics in a hospital is a complex process that deals with various items with different characteristics and requirement including pharmacy items (drugs, medical equipment, etc) and non pharmacy items (office equipments, foods and beverages, linens, etc). Hospital logistics management also involves various stakeholders with varying and often conflicting objectives. Hospital must spend a lot of resources to keep and maintain all the pharmacy and non pharmacy items. According to Brennan [2] healthcare supplies comprise 20 – 30% of the hospital’s total spending. This certainly becomes a burden to the hospital, which may indirectly affect the patients. In order to effectively and efficiently manage their logistics, the hospital managers must first understand the current logistics performance. Therefore, managers will need a tool that can display relevant information regarding logistics

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

A dashboard is also known as a manager’s dashboard, an executive cockpit, digital cockpit or a business scorecard. Bose [12] describes a dashboard as a software application that provides a single-screen display of relevant and critical business metrics and analytics to enable faster and more effective decision making. A dashboard provides a summary of the critical measurements needed to make the daily business decisions that affect an organization’s performance. A set of Key Performance Indicators (KPI) such as gross profit, inventory levels, the list of current top customers, etc becomes the basis of executive dashboard. Dashboard has been applied in various fields. Phipen et al [13] describes the use of monthly and weekly dashboards to report the performance of a large multinational airline company’s website including visits, visitors, registrations and visits to bookings information. Schulz and Heigh [14] reports the use of dashboard to show four indicators of regional logistics units of International Federation of Red Cross and Red Crescent Societies (IFRC). The indicators on this dashboard are displayed in various ways. Some are displayed

86

as monthly figures, some as year-to-date aggregation and some in a chronological sequence. According to Malik [15], the work on dashboard presentation can be divided into three areas: design, layout and navigation. He stated that a well-designed dashboard must appeal aesthatically and able to utilize a limited space to visualize a wealth of information. Some of the key elements for a good dashboard design are: 1) screen graphic and colors, 2) selection of appropriate chart types, 3) relevant animation, and 4) optimal content placement. The color pallette used should not distract from the key messages displayed on the dashboard. The next important aspect in dashboard design is the chart selection. Rasmussen et al [16] provides several guidelines in selecting the chart. Some of the most commonly used charts are as follow: - Area Chart is used to display trends over time or categories - Line Chart displays trends over time or categories - Pie Charts display the contribution of each value to a total - Clustered Column Chart compares values across categories Some dashboard is also built with animation which uses advanced visual capabilities (if provided by the software) to meaningfully interact with users. The final aspect to consider in dashboard design is content placement. The main principle is to limit the dashboard content to the most important KPI in order to avoid a sense of clutter that would overwhelm the user [15]. The design of dashboard screen layout has to consider three factors i.e. number of windows/frames, symmetry and proportions, and screen resolution. The final area in dashboard presentation is navigation. The whole information must be divided across different screens. It also involves linking charts and reports to allow user drill-down for greater details.

As shown in figure 1 this research is divided into three stages: intelligent phase, design phase, and implementation phase. The intelligence phase starts with literature study, survey, gathering objectives and KPIs (Key Performance Indicator) from management level, and collecting necessary data. The design phase consists of designing and constructing database for the collected data, defining presentation model for each KPI, and designing the storyboard. Designing the storyboard consists of conforming presentation model of each KPI to the current available technology, designing layout, determining dashboard titles and labels, and determining dashboard functionalities (drill down and/or alerts). The dashboard is developed in the implementation phase based on the design. Finally, the dashboard is reviewed for feedback. IV. INTELLIGENCE AND DESIGN PHASE This section describes the analysis and design processes in developing the dashboard. A. Intelligence Phase This research constructs and implements dashboard for measuring the performance of logistics process in hospital XYZ. Hospital XYZ is a government hospital located in Surabaya, Indonesia which has 44 units including inpatient, outpatient, emergency unit, etc. All these units obtained their pharmacy items from logistics unit. The logistics process in hospital XYZ is still conducted manually. The inventory data including item in and out are recorded in stock cards. The entire logistics process is evaluated quarterly. The dashboard developed in this research is an operational dashboard as it is focused on logistics unit in the hospital. The first step in developing dashboard is to understand the objectives and key performance indicators (KPI) for the user. Based on interviews with management, it was found that the main objective of the hospital is to provide high quality service to the patient in affordable cost. In order to support this, the logistics unit must provide the right items as needed by their internal customers. In other words, the main objective of the logistics unit is to provide high service level to their customers. Thus, they need a dashboard that would enable them to evaluate their service level. Based on the objective, three metrics and three Key Performance Indicators (KPI) can be derived. The metrics are: inventory level, inventory value and number of item in and out. The inventory level measures the average inventory that the hospital maintains over a certain period. The inventory value is an important metric to give an indication of the most valuable inventory in the hospital. Finally, item in and out show the turnover of the inventory. The first KPI is inventory stock out that measures the amount of items shortage over a certain period. The second KPI is inventory service level that measures the percentage of stock out occurrence over the number of times demand occur in a certain period. Inventory fill rate is the last KPI that

III. METHODOLOGY The methodology used in this research follows the three stages in developing business intelligence applications proposed by Turban [17].

Figure 1. Dashboard development methodology

87

inventory fill rate, the category menu will show two additional metrics i.e. inventory stock out and item in and out in the corresponding category within a certain period. More detailed information regarding two KPIs i.e. inventory service level and fill rate can be obtained through drill down functionalities. As shown in table 4, the pie chart of all categories in general menu or pie chart of all items in a category menu can be drilled down to show the value of each item in a certain period in the form of a table record. The inventory service level and inventory fill rate shown in general or category menu can also be drilled down for each item in a certain period. Alerts are given when inventory fill rate and inventory service level is below a certain threshold. Based on the information obtained from hospital XYZ, this threshold is set to 50% for both KPIs.

measures the percentage of stock out over the total demand. Presentation model for each indicator are then designed based on the characteristics of the indicators. Following the determination of objectives, metrics and KPI, data and information are gathered for the dashboard. Pharmacy items in hospital XYZ is divided into four categories i.e. medicine and herbs, medical instruments, reagent, and equipment. This research only focuses on the first three categories as the turnover for the last item is so low. In each of the three categories, several items noted by the managers as key items are used as samples for the dashboard. In the medicine and herbs category three items are taken as samples: alcohol, infus fluid PZ, and Invicloth. Spuit 3cc and Muslin rolls are two key items used in medical instruments. In reagent category two items are used: Combur M10 and Glucose. The information for each item are collected for two years from 2008 – 2009.

TABLE I DASHBOARD MENU AND DRILL DOWN FUNCTIONALITIES

B. Design Phase In the design phase, the presentation model of each KPI obtained from the intelligence phase is constructed. As suggested by Malik [15], the dashboard uses neutral color for the background and contrasting colors for the graphs. The dashboard is designed to have four menus that include one menu for general information and three menus for each category of items used in this research. The general menu shows the overall service level performance of the entire category over the two year periods while the category menu shows the service level performance of items within that category. Each menu shows the inventory value, inventory service level and inventory fill rate. Drill down functionality is provided for inventory service level and inventory fill rate in a certain period. The chart used is designed based on the characteristics of each metric. The inventory value is presented in a pie chart so that the user can see the contribution of each category or items over the total value. In terms of inventory service level and fill rate, managers should be able to generally see the trend over periods of time as well as compare the value for each item. Therefore, line charts are used to represent inventory service level and inventory fill rate in the general menu while column charts are used to display the KPIs in the drill down menu. In this research, the dashboard is designed to have multi dashboard screen. It consists of two rows and three columns to show three to five metrics and KPIs for each screen with proportion according to the order of importance. In order to avoid screen scroll which is not ideal for a dashboard usage, the screen resolution is set into 1024x768. Dashboard menus and drill down functionalities are summarized in table 1. The user is expected to view the general menu in order to obtain overall service level performance of the logistics. Then, from the general menu they can navigate to category menu for more detailed information about the performance of each category. User can also go directly to each category by choosing category menu and inputting the corresponding category and period. In addition to inventory values, inventory service level and

Information

General Menu

Menu for each category

Inventory values

all categories (pie chart)

all items in each category (pie chart)

Inventory service level

all categories (trend line chart)

all items in each category (trend line chart)

Inventory fill rate

all categories (trend line chart)

all items in each category (trend line chart)

Drill down

Each item (table record) Each item (column chart) Alert for critical level Each item (column chart) Alert for critical level

V. RESULT A. Implementation The presentation model, dashboard layout, and functionalities in the design phase are implemented using PHP. The application is tested to ensure that it works as planned. Results from the implementation are discussed in this section. Figure 2 shows the dashboard layout for the general menu. The first chart that appears in the left side is the pie chart that shows the contribution of each category over the total. Line charts for service level and inventory fill rate are displayed in the top right and bottom right of the screen, respectively. In these charts, the X axis represents periods of logistics (per quarter-year) from 2008 to 2009, while the Y axis represents the percentage from 0 to 100 percents. The dashboard in Figure 2 can be drilled down into items in each category. Figure 3a shows the total value of reagent category. Figure 3b shows the inventory service level for each item in reagent category. Since it shows the drilled down of reagent category, the average service level for each item

88

within category in each period should conform to service level in Figure 2 for the corresponding category.

Figure 3b. Inventory service level of each item in reagent category

Figure 2 Dashboard for all categories

For example, period P5 in Figure 3b has inventory service level at 100 and 0 for glucose and Combur M10, respectively. Summing up and averaging them will yield value 50 which is shown in Figure 2 for reagent category in period P5. This case is also prevailed for inventory fill rate shown in Figure 3c.

Figure 3c. Inventory fill rate of each item in reagent category

Figure 3a. Average of inventory values of each item in medicine and herbs category

Each period of inventory service level and inventory fill rate can be drilled down further. Figure 4 shows the drill down of reagent category in period 5. Alert tables are provided at the bottom of corresponding KPI charts. When the inventory service level or inventory fill rate fall below 50%, that item is printed in red and blinking. Two additional charts that shows the stock out and number of stock in versus stock out for reagent category within each period are displayed in the far right of the screen. A link to a table that shows each item values in reagent category within period P5 is provided in the bottom right of the screen.

Figure 4. Drill down of reagent category within period P5

B. The Use of The Dashboard Based on the implementation several points can be derived. Firstly, based on the general menu it can be seen that

89

the reagent category contribute 50% of the total value of all items used in the sample, which means that it is an important category. The performance of this category can be evaluated by observing the service level and fill rate trends. It was found that this category has a low performance where the service levels are only 50% in six out of 8 periods. In period 8, the service level even falls below 50%. Weak performance is also found in the fill rate. The chart shows that fill rate only reach 50% twice over the course of two years. These trends indicate problems which need to be investigated further. Further investigation on the reagent category through category menu shows that service level for Combur M10 in period 8 is 0, which means that the logistics department cannot provide this item when request comes. Still in the category menu, the item in and out chart shows that there are less Combur m10 item that comes into the logistics unit compared to demand for that item. The cause of such a bad performance need to be investigated so appropriate course of action can be identified and taken to improve the performance. Observation on the general menu reveals that medical equipment category also has a high contribution (41%) of the total inventory value. However, unlike the reagent category the medical equipment have relatively good performance. The fill rates of this category are always above 50%. The service levels are always above 50% except for period 5. The medicine and herbs category have the lowest contribution on the total value of inventory. This category also shows a weak performance where service level fall below 50% in period 6 and 7 and the fill rate is very low in period 7. Further investigation on medicine and herb category shows that service level and fill rate of alcohol, which belongs to this item is so low. In several periods, the service level and fill rate is 0. This certainly needs to be improved as the demand for alcohol is high. These findings suggested that the dashboard have served its purpose as a tool that enable hospital manager to evaluate service level of key items and quickly identify those that do not meet the targets. As the hospital currently has no other means to conduct such evaluation, the dashboard is regarded as a step forward to manage their logistics. However, several gaps need to be investigated further. Due to data availability, the KPI’s in this work are only those related to service level. This is fairly limited compared to previous work from Schulz and Heigh (2009) that include financial control, process adherence and innovation and learning as indicators. Thus, further work can add more indicators to enable wider analysis of the logistics management. There is also a need to add more detailed drilldown capabilities in the dashboard to enable managers to conduct more thorough root-cause analysis on the problems. This is an area that we would like to investigate further in our future work.

Implementation using several items as samples has demonstrated the potential use of the dashboard. It can show problem areas that the hospital can investigate further. It was found that the category that contributes highly to the total inventory value (50%) has performs badly in terms of service level. Further investigation shows that one item in this category has very low service level because the amount of item coming in is lower than the demand. The work presented here is still an initial effort to address the issues of hospital logistics management. The work can be continued by developing more detailed dashboard that capture activities in the logistics process so more specific problems can be identified and specific actions can be taken. REFERENCES [1]

[2] [3] [4]

[5]

[6]

[7] [8]

[9]

[10]

[11]

[12]

[13] [14]

[15] [16]

[17]

VI. CONCLUSION This paper presents the development of dashboard for evaluating logistics performance of hospital XYZ.

90

Tung F, Shang S and Chou C (2008) An extension of trust and TAM model with IDT in the adoption of the electronic logistics information system in HIS in the medical industry. international journal of medical informatics 77: 324–335 Brennan C D (1998) Integrating the healthcare supply chain, Healthcare Financial Management, Date: 1/1/1998 Côte M J (1999) Patient flow and resource utilization in an outpatient clinic. Socioeconomic Planning Sciences 33 (3):231–245. de Oliveira M (1999) 3D visual simulation platform for the project of anew hospital facility. In: De Angelis, V., Ricciardi, N., Storchi, G. (Eds.), Monitoring, Evaluating, Planning Health Services. Proceedings to the 24thmeeting of the ORAHS EURO-WG. World Scientific, Singapore, pp. 82–95. Swisher J R, Jacobson S H, Jun, J B & Balci, O (2001) Modeling and analyzing a physician clinic environment using discrete-event (visual) simulation. Computers and Operations Research 28 (2): 105–125. Blasak R E, Armel W S, Starks DW, Hayduk, M C (2003) The use of simulation to evaluate hospital operations between the emergency department and a medical telemetry unit. In: Proceedings of the 2003 Winter Simulation Conference, pp. 1887–1893 Sinreich D & Marmor Y (2005) Emergency department operations: The basis for developing a simulation tool. IIE Transactions 37: 233–245. Fletcher A & Worthington D (2007) What is a “Generic” Hospital Model?. Lancaster University Management School Working Paper, 2007/003. Burke E K, Causmaecker, P D, Berghe G V, and Landeghem, H V (2004) The State of the Art of Nurse Rostering. Journal of Scheduling 7: 441–499, 2004. Burke, E K, Decausmaecker, P, Petrovic, S, Berghe, G V (2006) Metaheuristics for Handling Time Interval Coverage Constraints in Nurse Scheduling. Applied Artificial Intelligence, Vol. 20 (9): 743-766. Aickelin, U, Burke, E K and Li, J (2007) An Estimation of Distribution Algorithm with Intelligent Local Search for Rule-based Nurse Rostering. Journal of the Operational Research Society 58: 1574 - 1585. Bose R (2006) Understanding management data systems for enterprise performance management. Industrial Management & Data Systems 106/1: 43-59 Phipen A, Sheppard L & Furnell S (2004) A practical evaluation of web analytics. Internet Research 14/ 4: 284-293 Schulz S F and Heigh I (2009) Logistics performance management in action within a humanitarian organization. Management Research News 32/11: 1038-1049 Malik S (2005) Enterprise dashboard: design and best practices for IT. John Wiley & Sons: New Jersey. Rasmussen N, Chen C Y & Bansal M (2009) Business Dashboard: a visual catalog for design and deployment. John Wiley & Sons: New Jersey. Turban E, Aronson, J E, Liang, T P [2007] Decision Support and Business Intelligence System. 8th Edition, Pearson Education International, 52-75.


Context-Aware Ubiquitous Musafir Dr.G.Subrahmanya VRK Rao+

Dr. Karthik Sundararaman+

Jinka Parthasarathi+

Prashant Parekh$ $

Retail Technology Consulting Group Cognizant Technology Solutions Chennai, India

+

Global Technology Office Cognizant Technology Solutions, Chennai, India

{subrahmanyavrk.rao, karthik.sundararaman, parthasarathi.jinka, prashant.parekh}@cognizant.com

application which runs on a Mobile Internet Device (MID) or in general, any Mobile device and would be able to connect to a Public Cloud at Real Time. „Musafir‟ enables an End User (a) A Context Aware Business Application that could assist Consumers for their Travel Needs. (b)To register Health Complaints and receive proper Diagnosis and Medicine (c) Access to Educational Videos (d) Demos of various Retail products to understand and estimate the quality of the products and the location they are available.

Abstract - Population using Mobile phones and Mobile applications is increasing every day. Emergence of Cloud Computing and advances in Wireless Communication technologies has changed the dynamics of mobile applications by providing Scalable On-Demand High Available Infrastructure at minimal cost with zero maintenance and empowered with faster data transfer. Enterprises have started to develop Cloud based mobile applications that can target their Customers who are on the move. This paper presents a Cloud based Context Aware mobile application that targets Travel, Healthcare, Education and Retail Enterprises and their Users. Machine Learning techniques such as Association rules were used to analyze the mobile end user data on the Cloud towards predicting the services preferred by the End User. I.

II.

A. Context Aware Travel Assistant Owing to the personalized nature of services, context aware applications could be developed to target personal travel [3]. Context aware applications are usually built by a set of manually defined rules which can define user behavior for varying context or user defined preferences. This could lead to inaccuracies and hence improper services. Machine learning empowered context aware application can improve the accuracy of such service discoveries. Bayesian network could be used to develop an intelligent engine, as this could address context aware problems by analyzing historical activity (behavioral log) of individual persons [4]. Reasoning could also be used to the study context in an ubiquitous environment [5]. The prominence of context aware application is steadily growing with the advent of mobile devices and other wearable wireless sensors. Indeed there was a lot of research literature available towards context awareness in an ubiquitous environment [6-8]. However when we talk about context awareness it mostly focuses on personalized services and it is well known that the number of end-users is also a constantly increasing factor. Scalability could well be an issue if the data collected for such an application is too large. „Cloud Computing‟ could be of help in such a scenarios.

INTRODUCTION

Marriage of Cloud computing and Emerging Wireless Technologies could ensure Pervasive Access and help towards Seamless Scalability, Zero Maintenance, Fault Tolerance, Minimal Infrastructure and On-Demand delivery of various Vital services such as Healthcare, Education and Shopping/Business experiences [1]. This has a tremendous impact on Information and Communication Technologies (ICT) which could eventually bridge the „Digital Divide‟ and specifically enables an innovative way to deliver ubiquitous services. We had developed a service platform entitled “Pervasive Cloud” on which various applications can be developed [2]. These applications could require high computing power, or could be database applications where huge amount of data has to be stored and accessed. Pervasive Cloud is based on convergence of emerging technologies such as Cloud and Mobile computing and could be used to provide inexpensive solutions with minimal infrastructure. Our Application „Musafir‟ which has been deployed on the “Pervasive Cloud” empowers the People-On-The-Move and provides context aware Travelling support, which at present is being extended to Education, Healthcare and Shopping applications. „Musafir‟ is a Rich Internet Mobile

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

MUSAFIR

Musafir employs Association Rules [9] towards Context Awareness, and could be used by End User/ Consumers or by Travel Service Providers to facilitate travel of their

91

customers. Association rules find association or relationships among large sets of data. These rules are Probabilistic in nature and provide results of Data Mining in the form of „if-then statements‟. Appropriate rules are selected from the set of all possible Rules, Constraints on various measures of Significance and Interest. Association rules have often been used to perform Market Basket Analysis successfully. In this study we had used Apriori Algorithm of the Association rules to determine Traveler‟s preference. We had used a part of the Public Mobile User Data set that was provided by the MIT media Lab [10], which consisted of User Profiles, Cell Tower ID and their Daily Routines etc. These details were stored in a Database which we hosted on a Public Cloud. In order to understand the usage of the said Datasets for a Travel based application, we had prepared a detailed travel record for each of the mobile user. The dataset we developed for user travel was divided into different categories viz. Very Frequent /Often/ Rare Travelers according to user Profile. Travel records consisted of their preferred mode of travel viz flight, train, bus or own car, journey date, their mobile and IMEI number. System context View of Musafir is shown in figure 1.

Fig. 2a. End User Form for Musafir‟s Travel Module

The End User clicks on the „Travel‟ button and fills in the details (see figure 2b).

Fig. 2b. Travel Service Form Fig. 1. System Context View of Musafir

End User‟s IMEI Number, Mobile Number, Time of Request and Place of request and other filled details would be sent to the JAVA Servlet running on the Cloud. Study was only concentrated on mobile phones whose IMEI number could be extracted [11]. Snapshot view of travel database used in the study is shown in figure 2c. The Servlet first executes SQL queries to extract History details about the Mobile user based on IMEI and Mobile Number. A temporary table, which consists of travelling patterns of a particular person, is formed and Aprori algorithm is applied on the resultant temporary table. The algorithm will extract rules based on the history details and would provide probabilities for mode of travel of a particular person as

Client End Form for the Mobile End User was developed using JAVA / J2ME for Services, which have Push Buttons for Services, an End User may require (figure 2a).

92

illustrated in figure 2d. Mode of travel that has the highest probability would be chosen and send to the person who initiates the travel request. Mobile screen shot of the response to the travel request is shown in figure 2e.

Fig. 2c. Snapshot view of travel record database Fig. 2d. Results of running Apriori Algorithm on the travel database for a particular person

93

provide certain services like Map, RSS, product search etc as API (Application Programming Interface) which can be invoked by general users. These API‟s could be very useful for a traveler. For example a Map service can help to locate a place or provide driving directions from a place to another. In Musafir we had used the Map Service API [12] to help the end user by providing the Driving Direction (from source to destination) if they opt for driving their own Transport / Car (figure 2g). Musafir provides a separate servlet to address this. This servlet has two functions viz. (a) map service API (b) API function provided by a 3rd party bulk Short Message Service (SMS) service provider. Once the end user indicates to the servlet that they prefer driving their own transport, Map Service API is invoked to get the driving directions for the travel and the link to page is sent as an SMS using the SMS API function to the users mobile phone.

Fig. 2e. Response from Servlet on the Cloud after processing End User request

End User would then be able to view his choice on their mobile screen. Once the End User confirms the Mode of Travel, a list of schedules of Preferred Mode of Travel from Source to Destination with all the details is retrieved and shown on the Mobile Screen of the End User (figure 2f). End User can then book for the Travel.

Fig. 2g. Map showing the Driving Direction between the Source and Destination

The two Servlets mentioned above are actually connected to a MySQL database table called „journeydet‟ present in the Cloud. Once the user confirms the bookings travel details of the users along with their mobile number, IMEI number, date of journey, source and destination of journey, mode of journey etc are stored in a table „journeydet‟. This will help in eliminating static rules and the dataset for Apriori algorithm will have an improved support such that it can provide more accurate results based on the Dynamic Database details. A small automated program (procedure), would run at the start of everyday. It has three main functions (a) search function, which is basically a SQL query to select details (Mobile number, source and destination of travel) of those who are travelling on that particular day from journeydet table (b) Invoke weather service API [13] and find out the respective weather conditions for each of the identified

Fig. 2f. List of Flights from source to destination with details of Discounts

One of the main objectives of context aware applications is to discover service according to the user‟s preference. Web has transformed so much that many enterprises

94

destination (c) SMS API which would then send the respective weather information to the corresponding mobile numbers. This information would help to facilitate the travel arrangements of the end user.

Education Module of Musafir empowers an End User to select and watch Videos relating subjects of their interest. The typical End User could be any consumer and specifically academic community around the world. The videos will be streamed from the Cloud (figure 4). Performance of such an application depends on bandwidth of the network and type of communication technology used. In our lab environment we were able to upload a 5MB file to the Cloud in roughly around 2minute 30 seconds to 3 minutes and the time taken to download the file / stream the file from the same is around 1 minute 30 seconds to 2 minutes considering the connection speed to be a maximum of 512KBPS DSL Connection.

B. Healthcare, Education and Shopping Modules End User forms (figure 3a) were developed based on Flex Technology which would facilitate Rich User Experience. User can fill in the Health Complaints and submit the details to the Cloud. A third party API which facilitates SMS to a Mobile device was integrated with the Servlet program. Once the Database (on the Cloud) is updated with the Health complaints, the API can trigger an alert to a registered expert Physician.

Fig. 4. Educational Videos that could be accessed by a Touch Screen interface

Fig. 3a. End User entering the health complaints

Shopping module of Musafir enables an End User / Consumer, to watch Demos related to various Retail Products, to understand the Location of their availability, their Price and to store their personalized shopping choices in a „Shopping Cart‟, which could connect to a Billing/Credit Card System (figure 5).

Physicians were provided with a mobile application, which would help them to assess the Patient‟s Health Complaints from the Cloud database [2]. Physician, up on analyzing the case, can submit their Diagnosis and Prescription details to the Cloud Database seamlessly, which would send an SMS alert to the patient who filled in the Complaint. This application also empowers the End User to transfer ECG and images to the Cloud which could also be viewed by the Physician (figure 3b).

Fig. 5. Demos of Retail products available at a particular location

Security is a major concern for web based applications, where there is a possibility of a large number of users. Having realized the sensitivity of datasets we are trying to address security issue by using 128 byte Advanced Encryption Standard (AES) algorithm on the datasets [14].

Fig. 3b. Physician‟s view of the ECG signal of a Patient retrieved from the Cloud

95

For each application there would a set of administrators and end users. Hence a role based access to the database is also being studied and implemented. We are currently working towards making Healthcare, Education and Shopping modules more „intelligent‟ by making them ContextAware. III.

[5] Donghai Guan, Weiwei Yuan, Seong Jin Cho, Andrey Gavrilov, Young-Koo Lee, and Sungyoung Lee* “Devising a Context Selection-Based Reasoning Engine for ContextAware Ubiquitous Computing Middleware” UIC 2007, LNCS 4611, pp. 849–857, 2007. [6] Dey, A.K., Abowd, G.D., Salber, D.: A Conceptual Framework and a Toolkit for Supporting the Rapid Prototyping of Context-Aware Applications. J. of HumanComputer Interaction (HCI) 16, 97–166 (2001) [7] Hong, J.I., Landay, J.A.: An Infrastructure Approach to Context-Aware Computing. J. Human-Computer Interaction (HCI), London, UK, 287–303 (2001) [8] Shafer, S.A.N., Brumitt, B., Cadiz,J.J.: Interaction Issues in Context-Aware Interactive Environments. J. HumanComputer Interaction (HCI), London, UK, 363–378 (2001) [9] Zijian Zheng, Ron Kohavi, Llew Mason, “Real world performance of association rule algorithms”, Proceedings of the seventh ACM SIGKDD international conference on Knowledge discovery and data mining San Francisco, California, Pages: 401 – 406,Year of Publication: 2001,ISBN:1-58113-391-X [10] N. Eagle, A. Pentland, and D. Lazer (2009), "Inferring Social Network Structure using Mobile Phone Data", Proceedings of the National Academy of Sciences, 106(36), pp. 15274-15278 [11] http://mobilepit.com/10/how-to-get-imei-number-in-j2mejavame.html [12] http://code.google.com/apis/maps/documentation/examples/d irections-simple.html [13] http://weather.weatherbug.com/desktop- weather/api.html [14] Christof Paar, Jan Pelzl, "The Advanced Encryption Standard", Chapter 4 of "Understanding Cryptography, A Textbook for Students and Practitioners". Springer, 2009.

CONCLUSION

Musafir was targeted towards Healthcare, Retail, Travel, Education /Content Management Business Groups and its Users. The application provides (a) Pervasive learning access through videos and On-Demand Education (b) Empowers Online Shopping and Business On-the-Move (c) Real Time and Pervasive Primary Health Care Service (d) Context aware Travel Solution. IV. REFERENCES

[1] G Subrahmanya VRK Rao, Jinka Parthasarathi, S. Karthik and GVN Appa Rao, “Implementation of Virtulization Oriented Architecture – A Healthcare Case Study”. IT Revolutions 2008, LNICST 11 proceedings, Springer. [2] G Subrahmanya VRK Rao,Karthik Sundararaman,Jinka Parthasarathi, Dhatri – A Pervasive Cloud Initiative for Primary Healthcare Services – Proceedings of IEEE-ICIN 2010, Berlin. [3] Cheverst, K., Davies, N., Mitchell, K., Friday, A., Efstratiou, C.: Developing a contextaware electronic tourist guide: some issues and experiences. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (The Hague, The Netherlands, April 01 - 06, 2000). CHI '00. ACM Press. (2002). [4] Masaki Matsudaira, Kyoko Hoshikawa, Kohei Taki “Context Awareness and Its Applications”, Oki Technical Review, April 2008/Issue 212 Vol.75 No.1,94-97.

96


An Information provision system based on a Multi-Hop RFID scheme for ITS (Intelligent Transportation System) Hiroaki Togashi

Shigeki Yamada

The Graduate University for Advanced Studies 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo, Japan

National Institute of Informatics 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo, Japan

Abstract- The information provision systems in ITSs (Intelligent Transportation Systems) currently use licensed bandwidth (e.g, VICS, DSRC), and requires a large-scale client-server system to gather “probe information”. The authors are researching an information provision system based on RFID schemes, aiming to make the system inexpensive to construct. One problem is that the basic RFID technology does not have enough communication distance for this purpose. In this paper, the authors propose an information provision system based on a Multi-Hop RFID scheme in order to overcome this problem. A Multi-Hop RFID scheme also enables information provision systems to provide “probe information” since they can exchange their information with each other at every hop. Since several types of information are to be handled, the proposed system adopts a hybrid transmission method that combines infrequent periodic transmission and transmission at information update. Although some problems still remain, the proposed system is advantageous in several ways. For example, it can provide traffic congestion information in real time without requiring a large server-client system. Keywords; ITS (Intelligent Transportation System), RFID (Radio Frequency IDentification), Prove Information, Road-to-Vehicle communication, Multi-Hop communication

pedestrians at low visibility crossroads, 2) providing color of traffic signal linked to navigation systems, 3) collecting and providing traffic congestion information, and 4) providing information of encountering emergency vehicles. The rest of this paper is organized as follows. Section II shows the current situation with respect to providing information. The information types handled in information provision systems are categorized in Section III. We describe our proposed system in Section IV and issues on the proposed system and possible countermeasures against them are shown in Section V. After that, we overview the ways in which the system can be applied and examine specific issues of each scenario in Section VI. Finally, we summarize this paper and touch on our future plans in Section VII. II. RELATED T ECHNOLOGIES A. VICS[2] In Japan, VICS (Vehicle Information and Communication System) is the main ITS scheme for information provision system. A problem with VICS is that although its data format can only handle 16 types of information, 7 types of it have already been handled; we can define only 9 new types of information from now. Another problem is that VICS is a one-way broadcast system; and an in-car transmission system is required to enable a vehicle to send information. B. DSRC[3] The DSRC (Dedicated Short Range Communication) enables two-way and wide-band communications, but its communication distance is short. This means that another large-scale server-client system is required when an information providing system wants to collect probe information (e.g. congestion information). C. Multi-Hop RFID[4][5] “Active-type RFIDs” that are capable of multi-hop communications are called “Multi-Hop RFIDs”. Information written in these tags is exchanged by using the well-known “bucket brigade” strategy. This feature makes it possible to collect information from outside the communication distance of RFIDs and transmit it to a vehicle. This strategy is especially suitable for collecting and providing probe information [1], and it

I. INTRODUCTION Currently, most vehicle information provision systems in ITS, such as VICS and DSRC, utilize licensed radio wavelengths. These information provision systems also require a server-client system in order to collect probe information[ 1 ], and this requirement makes them large-scale one and costly to maintain. The authors are researching an information provision system based on RFID scheme to solve these issues, and found that the communication distance between RFIDs and Reader-Writers (R/W) that can be obtained with current RFID technology is not sufficiently long for this purpose. This paper describes an information provision system based on a Multi-Hop RFID scheme as a means of solving this short range problem. We evaluated the characteristics of our proposed system by considering the following application cases: 1) detecting vehicles and

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

97

receiver will likely be large. In this situation, an information provision system needs need an identifier to enable it to recognize the source of information. The vehicle may also receive the information about the color of signals that are not on the vehicular route, but the vehicle should ignore such information. (c) Intermediate (IM) Information This information is similar to RT/QRT information in the sense of sending and receiving receiv information if the information has been updated. However, IM information must be sent periodically because its validity period is much longer. On the other hand, this is unnecessary if the system provides RT Information, QRT Information, and IM information at the same time. Thus, there should be only one sender each time information is updated. (d) Semi-static (SST) Information This information is similar to IM information, information except that its validity period is longer and periodic transmission thus become more important. When this type of information is updated, update there should be only one sender. Furthermore, some of the SST information with a long update interval will be already-known information, in the same way as ST Information. (e) Static (ST) Information This information does not change frequently. It will be subject to at most a few updates update and the system can rely on periodical transmission. The system would be more useful if system could ould collect information from more remote distances. In any event, the only useful information in this case is the one pertaining to the vehicular route and other information should be discarded. Moreover, since most of the information is already known, the system should effectively complement it. IV. PROPOSED ROPOSE SYSTEM Our proposed system utilizes a Multi-Hop RFID scheme. We review our proposed roposed system in this section. A. Information Provision Method

eliminates the need for huge server-client client systems. Our proposed system employs “Multi-Hop Hop RFIDs” RFIDs as a platform for collecting and providing ding information by making use of these characteristics,, because these RFIDs are suitable for providing information over wide areas in real-time. III. DEFINING OF INFORMATION TYPES PROCESSED IN INFORMATION PROVISION SYSTEM

Before showing details of our proposed system, we here define information types in order to clarify the requirements of information provision system. A. Classification of information types provided in information provision systems Before designing our system, we needed to classify the information provided in information provision systems. Table 1 shows a classification focusing on the time intervalss at which the information was updated. This table shows that information with a short update interval should be transmitted tted whenever it is updated, and information with a long update interval should be transmitted periodically. T ABLE 1 CLASSIFICATION BASED ON THE INFORMATION TION UPDATE INTERVAL INTERV Category R e a l - t i m e (RT) Update Interval

~Sec.

Q u a s i I n t e r - Semi emi static stati real-time mediate (QRT)

(IM)

Sec.~

Min.~

Min.

Hour

(ST)

Hour Day Hour~

Day~

T y p e o f 1. Pedestrian Color of T r a f f i c R information presence info. provided

2. Vehicle

Static

(SST)

o

a

d 1. Name

traffic

accident

Maintenance

of place

signal

info.

info.

2. Traffic

presence info.

regulation info.

Transmission

Transmission on

Infrequent

method

information update

periodic eriodic transmission

B. Information characteristics This section outliness the characteristics character of the classified information. Here, “sender” means the source of the information (e.g., pedestrian, traffic signal etc.) and “receiver” means the destination of the information (e.g., vehicle etc.). (a) Real-time (RT) Information This information must have a very short update interval. If there is only one sender at a time, time tags can send information only if the information is to be updated. Here, the receiver receives only that information. However, a single sender does not allow collecting all essential information for the system.. For example, a single sender cannot make decisions regarding traffic congestion, and furthermore, it is unable to determine the presence of pedestrians. These issues can be resolved when there are many senders, but we should consider congestion on the signaling path in that case. (b) Quasi real-time (QRT) Information This information has characteristics similar to those of the RT Information. Especially in the case of traffic signals, the vehicle must collect information from a long distance and the number of hops from sender to

Fig. 1 Information provision image of the proposed system

Figure 1 shows how information is transmitted and exchanged in the proposed system. system Figure 2 shows the system components relevant evant to the information provision between a vehicle and RFIDs. From the vehicle viewpoint, an active ctive RFID broadcasts several

98

considering how to recognize the lane where the vehicle is running. We chose the “Boundary Scheme” illustrated in Fig. 4 because;; 1) 1 this layout is easy to be utilized for multiple purpose, purpose and 2) this layout enables a vehicle to recognize the lane where it is running.

types of information with its ID and the on-car Reader-Writers (R/W) merely receives it. When to transmit information from an RFID differs according to the information types shown in Section on III.B, and there are two types of information; information ation that requires infrequent periodic transmission and information that requires transmission whenever information is updated. Considering that transmission is not always performed within a certain duration when information is updated, a hybrid method of infrequent periodic transmission and transmission at information update must be able to provide information in real-time. From the RFID viewpoint, information should be written to RFIDs as follows. First, an R/W that notices notice a state change (e.g.,, color of signal changed from red to green) updates the information in RFIDs around it. After that, this information is delivered to other RFIDs by the bucket brigade strategy. RFIDs FIDs also exchange their information at infrequent periodical transmission transmis in the same manner as in the method of information update. In this way, information can be delivered over the wide areas.

Fig. 3 Road-surface surface Scheme (for Vertical placement)

Fig. 4 Boundary Scheme (for 2-D placement)

V.

ISSUES ON THE PROPOSED SYSTEM SYST AND POSSIBLE COUNTERMEASURES AGAINST THEM

In this section, we show issues on the proposed system and their possible countermeasures against them. Generally speaking, there are three t main issues remaining to be solved. The first f one is how the vehicle can recognize whether an RFID is placed on the left side or right side. The second one is how to choose information that will be useful usefu or meaningful for the vehicle. The he last one is that data congestion will become a major problem when many vehicles (or R/Ws) want to write to a certain RFID.

Fig. 2 Components of the proposed system

B. Equipment Placement We assume that the RFID scheme applied in our proposed system is an Active-type type Multi-Hop Multi RFID scheme, and that the layout of RFID tags and Reader-Writers (R/W) is as follows. Vehicle/roadside equipment is placed as in the “On-Car system” defined in [ 6 ].. Who has R/Ws is defined by who will be the source of information. For example, Section III shows that pedestrians will have R/Ws for pedestrian existence information ation and traffic lights should be equipped with R/Ws for sending color of traffic signal. In considering the layout of RFIDs, we should choose one of the schemes from vertical placement and 2-D placement as described in [6]. Simply saying, s in the former case, layouts of RFIDs are classified according to whether RFIDs are placed above vehicles or below vehicles. We choose the “Road-surface surface Scheme” Scheme from these layouts, illustrated in Fig. 3. And in the latter case, layouts are classified according to whether RFIDs are placed on the border of laness or the center of lanes. lanes However, which of them should be applied is defined by

Fig. 5 Concept and possible usage of RFID-map RFID

(a) Recognizing whether an RFID is placed on the left side or right side of vehicle In order to solve this issue, our o system includes an RFID-map map that plots its position and ID, inside the vehicle. Also, vehicle can detect its velocity and going direction. This information enables the vehicle to estimate that the RFID is on the right or on the left le side. A problem relative to updating the RFID-map still remains; one solution to resolve it is to use the same technique as that currently used to update car-navigation systems.

99

Multi-Hop Hop RFID scheme enables simple system construction. Although Multi-Hop Multi RFID schemes have a power supply problem, our proposed system is still more effective than the existing systems because it fully utilizes RFID characteristicss and it is simply achieved by Multi-Hop Hop RFID scheme.

(b) Choosing useful or meaningful information for the vehicle In the basic Multi-Hop Hop RFID scheme, there t are no limitations as to where RFIDs should transmit information. Useful seful information differs between each vehicle and therefore it is impossible to set up specific constraints. Consequently, a possible countermeasure against this issue is for the vehicle to choose information that would be important for it; it in this case, the RFID-map is useful as shown in Fig. 5. (c) Data congestion The last ast one is that data congestion becomes a major problem when many vehicles (or R/Ws) want to write to a certain RFID. This problem is related to “Active-type RFIDs” and various countermeasures against it can be considered. One possible solution is to designate target RFIDs when an R/W tries ies to write information to them. them Non-designated RFIDs merely ignore this written information and if an R/W fails to designate target RFID(s), no information can be written to the RFIDs. (d) Security and privacy related issues We also understand that we should consider security issues and main issues are related to information writing. In our proposed system, anyone can read information written in RFIDs and anyone can write information to any RFIDs. However, we must avoid illegal illega information writing and ensure the integrity of data written in RFIDs. Consequently, important information requires strict authentication and authorization while noncritical information does not need them.. For example, emergency vehicle encounter information tion must not be forged by malicious people because such forgery fo is socially inappropriate.. On the other hand, pedestrian detection information is only used for information providing and misinformation tion only causes little damage. damage From a viewpoint of privacy,, we should store information in RFIDs as anonymous information, information in order to avoid illegal tracking of vehicles and pedestrians. In addition,, IDs related to vehicles vehicle and pedestrians should be used only in the first hop (i.e., communication between “sender” and RFID), while anonymized IDs should ld be transmitted in later hops (i.e., communication among RFIDs or between RFID and “receiver”). VI. APPLICATION SCENARIOS AND SPECIFIC SPECIF ISSUES A. Detecting vehicless and pedestrians at low visibility crossroads (a) Overview of the system behavior In the proposed system, we should reduce underlying risks by providing the vehicular presence information or pedestrian’s presence information shown in Fig. 6. In this situation, vehicles write vehicular presence information for pedestrians and other vehicles, and pedestrians write pedestrian presence information for vehicles with their Reader-Writers (R/W R/W). The proposed method has two new features related to utilizing RFIDs: 1) it can send information to pedestrians and simultaneously detect a pedestrian’s pedestrian presence because pedestrians have R/W; R/ and 2) the

Fig. 6 Providing reminders in proposed system

(b) Issues and countermeasures One important issue is how to determine whether the information is coming from the right or the left side of the vehicle. This problem is resolved with minor revision of the countermeasure shown in Section V. Any determination should take into account the positional relationship between betw the ID that is initially written information and the ID that the vehicle receives information from. Note ote that a position estimation method that uses only the absolute position (e.g., longitude and latitude) of a vehicle may fail to recognize the vehicular going direction. Another solution is to equip the vehicle with receivers on both sides and identify which receiver got the information. However, this method is difficult to apply because information from the right side is not always received by the right-side receiver. B. Providing color of traffic signal linked to navigation systems

Fig. 7 Providing color of signal in proposed system (a) Overview of the system behavior The proposed system provides color of signal to drivers as shown in Fig. 7. 7 In order to help drivers recognize which color of traffic signal which drivers

100

expect to encounter along the navigated route, route our system provides the current color of signal to the vehicle and shows it on the vehicle navigation system. Our system also settles communication distance problems by applying the bucket brigade strategy. (b) Issues and countermeasures One issue in this case is how to provide information if no route is given in the navigation system. Even in this case, our system can provide information by appending directional information on the color of signals † , illustrated as Fig. 8.

Our system can collect and provide prov this information as shown in Fig. 9. A Multi-Hop Multi RFID scheme can collect information at every hop. This feature enables the system to recognize a vehicle’s vehicle real-time situation and suggest how the driver could change the route to avoid congestion. The receiver can also recognize how vehicles are stopping,, based on the sender IDs and put this information on the RFID-map RFID shown as Fig. 5. This RFID map plots its position and ID, inside insid the vehicle. It also so becomes possible to use other information such as the color of signals to know the reasons of the congestion. (b) Issues and countermeasures One issue in this case is that the system may sometimes misinterpret the fact that temporarily stopped cars (e.g., those waiting for a red signal to change to green) may signify a state of congestion, whereas the system actually has only recognized the fact that a certain vehicle has stopped. Another issue is that the system may provide information for a certain location even if the driver has no intention of going there. In this case, the receiver should simply ignore such information. nformation.

Fig. 8 Providing color of signal with directional information

Here, we can consider two ways in providing directional information; 1) names of place where a vehicle will arrive there when it travels on the route and: 2) direction represented as north, southeast etc.. Also, there are two kinds of drivers; they are familiar with the location around the crossroad and they are not familiar with it. Consequently, it is more useful that tha the system provides both of directional information. Also, the information can be provided by voice navigation or on-screen screen navigation, but the former is more suitable for the driver. C. Collecting and providing traffic congestion information (a) Overview of the system behavior

Fig. 10 Communication distance of an RFID and amount of information

Too much amount of information is another problem. Suppose uppose that the writing interval is one second. In this case, considering the communication distance, there are about four vehicles that can use a single RFID tag. Consequently, four-traffic traffic congestion information is written to a certain RFID tag within one second. Thus, the amount of congestion information is easy to increase and we should consider some countermeasures against it. One possible countermeasure against this problem is to reserve only one area for traffic congestion information. The he amount of traffic congestion information can be reduced and it is updated in the same way as in usual overwriting cases. To recognize traffic congestion situation, only one congestion information foe each RFID is still useful; a vehicle can recognize congestion situation of each RFID from its information and estimate the situation over a wide area using their information.

Fig. 9 Providing congestion information

†

Pronunciation symbols are enough to describe them because main

purpose of it is for voice navigation

101

D. Providing emergency vehicle information (a) Overview of the system behavior

VII. CONCLUSION We have proposed an information providing system based on a Multi-Hop op RFID scheme. s By utilizing this scheme, short range problems problem that the RFID-based systems encounter were overcome. Furthermore, the bucket brigade strategy, one of the characteristics of the Multi-Hop RFID scheme,, makes our system possible to provide and collect real-time time information for vehicles. In evaluating the proposed system, we considered and showed issues to be solved. As a transmission method, the proposed system adopts a hybrid method of periodical transmission and transmission tra when information is updated,, because several types of information are to be handled. handled Although some issues still remain to be solved,, the proposed system has many advantages. For example in providing provi traffic congestion information, it allows real-time real information to be provided to vehicles without any huge server-client system. This paper only describes the concept of our research and our ongoing work is as follows. Currently, we are doing research on “Detecting Detecting vehicles and pedestrians at low visibility crossroads”. crossroads We are conducting simulations in some road/pedestrian existence situations and evaluate; 1) travel time from the source of information to RFIDs placed on a certain distance and 2) how much memory is necessary in RFIDs (peek, average, median etc.). As subjects for future work, we will consider each application cation case and attempt to solve the issues that still remain. The main m issues are 1) how to recognize whether an RFID is placed on the left or right side of the vehicle, 2) how to choose useful information for the vehicle, vehicle and 3) security related issues especially related to data writing. writing We will report solutions for these issues on another occasion.

encounter

Fig. 11 Providing encountering information in proposed system

As Fig. 11 shows, our system can provide emergency vehicle encounter information by utilizing a Multi-Hop Multi RFID scheme. Our system can recognize the presence of emergency vehicles even if the driver cannot cann see or hear them. Thus, the system can help emergency vehicles to reserve driving lanes after they make a turn. The system can also help drivers to recognize that they are on the route of an emergency vehicle by adding the emergency vehicle’s route to this encounter information. This can alert drivers who may tend to wander or make unnecessary avoidance maneuvers. (b) Issues and countermeasures

REFERENCES [1] M. Sarvi, et al., “A A methodology to identify traffic condition using intelligent probe vehicles” vehicles proceedings of 10th World Congress on Intelligent Transport Systems, Systems 2003. [ 2 ] T. Daimon, et al., “Study Study on Safety Assist Information of Advanced Cruise-Assist Assist Highway Systems (AHS) using VICS in Blind Curve Section of Urban Expressway” Expressway Journal of Mechanical Systems for Transportation and Logistics, Logistics Vol. 1 , 2006, No. 2 Special issue on Driving ng Simulation Conference-Asia/Pacific Conference 2006, Fig. 12 Possible solution; providing the route of emergency

pp.192-202.

vehicle with encounter information

[3] S. Olariu and M. C. Weigle, “Vehicular Vehicular Networks: From Theory

We should determine the best way to indicate the emergency vehicle routes for a driver to know whether the emergency vehicle is approaching his/her vehicle or not. One solution is to provide an indication of where to turn right or left at some distance from the RFID that was written the encounter information, shown in Fig. 12. With this information, the receiver can recognize if an emergency vehicle is moving towards the driver’ss vehicle or not. Here, our proposed system utilizes RFID scheme. Consequently, where to turn right or left should be represented by ID of an RFID that is placed close to there.

to Practice” Chapman & Hall, 2009. 2009 [4] Y. Yoo, et al., “RFID RFID Reader and Tag Multi-hop Multi Communication for Port Logistics” IEEE International Symposium on a World of Wireless, Mobile and Multimedia Networks (WoWMoM 2009), 2009, pp.1-8. [5] S. H. Hong, et al., “ISO/IEC ISO/IEC 18000-7 18000 based on RFID multi-hop relay system” Proceedings of the 9th international conference on Communications and information technologies, technologies 2009, pp.1450-1454. [6] H. Togashi and S. Yamada, Study on Detection of Car's Location using RFIDs and Examination of its Applications, Applications IPSJ SIG Technical Reports, 2008-ITS-57, 2008, pp57-61.

102


A Semantic Index Structure for Integrating OGC Services in a Spatial Search Engine J. Márquez (1), J.E. Córcoles (2), A. Quintanilla (3) (1) SIGTEL Geomática. [email protected] (2) LoUISE Research Group. [email protected] (3) Remote Sensing and GIS Group. [email protected] Castilla-La Mancha University s/n.02071.Albacete. Spain +34967599200 Abstract- There are a huge number of Spatial Data Infrastructures (SDIs). This has several advantages, but it is really difficult for a user to know what spatial service could satisfy his/her needs. For this reason, the SDI community now demands an approach to integrate SDIs and relate them with semantic features. In order to contribute to a solution, in this paper we propose an approach to construct an ad-hoc semantic-index structure (ontology) to relate political divisions, non-political place names and layers offered by OGC services. In this way we can integrate spatial OGC services with semantic concepts. Then, a search engine allows users to find OGC services using semantic criteria. Keywords: Semantic index, OGC, IR, GSD, spatial information.

I. INTRODUCTION In the last decade, the research community has begun an effort to investigate foundations for the next stage of the Web, called the Semantic Web [1]. Within the Semantic Web, a rich domain that requires special attention is the Geospatial Semantic Web [2]. The enormous variety of encoding of geospatial semantics makes it particularly challenging to process requests for geospatial information. In the near future, the Geospatial Semantic Web will allow the returning of both spatial and non-spatial resources or services to simple queries, using a browser [2]. However, in the same way as with the Semantic Web, in order to approach the Semantic Geospatial Web it is necessary to solve several problems. One of them is the addition of semantic to spatial information and the definition of structures to carry out efficient queries over this information or services. In order to do this, the solution has to be framed in two topics: Information Retrieval (IR) and Geospatial Services Discovery (GSD) [11], which deals with methods for automatically searching for geospatial services published on the Internet. In this paper we propose an approach to construct an ad-hoc semantic-index structure (ontology) to relate political divisions, non-political place names and layers offered by Open Geospatial Consortium (OGC) services [20]. In this way we can integrate spatial OGC services with semantic concepts. Then, a search engine allows the user to find OGC services using semantic criteria.

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

103

In our approach, using the keyword entered by a user in the browser, the system returns the spatial services (previously crawled) that satisfy the keywords. In order to do this, we use political divisions and non-political place names (alphanumeric semantic information) to discriminate the result. The relation between the information provided by the spatial services and its alphanumeric metadata (political divisions) is represented by an ontology. The capability of automating each process of setting up and updating the system is one of the most outstanding features of this project. Spatial services are standardized by the Open Geospatial Consortium (OGC) in order to improve the interoperability between different geospatial systems. OGC services are the basis of Spatial Data Infrastructures (SDIs), which are frameworks of spatial data, metadata, users and tools that are interactively connected in order to use spatial data in an efficient and flexible way [18]. In addition, this project has taken into account many common problems in geospatial search engines. Thus, it provides the spatial operator contains between political geometries, and semantic relationships such as meronymy (part-of) and hyponymy (subsets) between terms belonging to the same knowledge domain. II. RELATED WORK Geospatial Web semantic architectures have been extensively studied in the literature. [3], [4] and [6] stand out. Boucelma et al., [3] present a mediation system that addresses the integration of GIS data and tools, following a global-as-view approach. It has a multi-tier client-server architecture based on Web Feature Service (WFS) [15] and uses standard wrappers to access data, extended by derived wrappers that capture additional query capabilities. Gupta et al., [4] extend the MIX wrapper-mediator architecture for integrating information from spatial information systems and searchable databases of geo-referenced imagery. MIX is focused on integrating georeferenced imagery but our approach is focused on spatial geometries. On the other hand, Córcoles et al., [6] designed a novel approach for integrating Geographic Markup Language (GML) resources. The proposed architecture uses a Catalog expressed by Resource Description Framework (RDF) to relate the GML resources. Although [6] has the same aims as [5], it has a different focus.

Focusing on IR, several outstanding studies have related textual references to geographic locations. C. Jones et al., [7] and M.B Lieberman et al. [8] have developed spatio-textual indexes to make the queries more efficient. Furthermore, a large number of alternatives are available in the literature to tackle the problem of representing spatial relationships while adding semantic features: Abdelmoty et al., [9] propose frameworks combining rules and Ontology Web Language (OWL) for managing place ontologies, while [10] defends using an R-Tree index held by a hash table for representing political divisions. A comparative between OWL and GML is given in [21]. On Geospatial Services Discovery - GSD (IR discipline), N. Chen et al., [11] describe crawling OGC services such as Web Map Service (WMS) [14] on the Internet and detecting which of them are really valid using service capabilities and checking the requested XML metadata document . Other proposals such as [12] make use of a service capability document to evaluate similarities between different services through an algorithm for semantic matching. With respect to geospatial semantics, there are many articles covering the problem of representing specific knowledge domains, in line with our project. Hochmair et al., [13] set out a solution using a taxonomy for every knowledge domain where every node represents a concept in reality and it is extended by all its synonyms. Meronymy and hyponymy between concepts are represented with the branches of such a taxonomy. In section III an introduction to our search engine architecture is given. Details about constructing a semantic-spatial index are shown in section IV. Section V shows how different scenarios use the semantic-spatial index in order to satisfy user queries. Finally, our conclusions and future work are given in section VI. III. REFERENCE ARCHITECTURE In order to integrate OGC services with semantic concepts in a search engine, we use a classical architecture basis with three tiers, as shown in Fig. 1. (1) The first tier (Internet) covers all OGC Services published on the Internet, besides some geo-name databases. Currently, we are interested in a subgroup of those geospatial services: WMS and WFS. This choice is based on statistics related to the most commonly implemented services in SDI. WMS is capable of retrieving map images in formats such as JPG, GIF or PNG. Moreover, WFS has an operation for retrieving features in formats such as GML and another related standard called WFS-G has been defined that is used as a Gazetteer [17]. Furthermore, place name databases such as Geonames [16] will be used to validate any political division besides getting the ascendant hierarchy of its political divisions.

104

(2) The second tier (Crawler) defines the main contribution of our approach in this paper. It has several elements for discovering, analyzing and integrating OGC services with semantic concepts. The main elements of this tier are: A. OGC Services Crawler It finds links on the Internet that are candidates for being an OGC service URL in the same way as [11]. Our contribution is that the first pages to browse are the first results obtained through different search engines (Google and Yahoo in our prototype) using keywords such as “SDI”, “GIS”, “OGC”, “list”, “WMS” and “WFS”. Once those candidates are found, they are checked by requesting a GetCapabilities WMS operation, or similar, depending on the specific service. The returned XML documents are stored in a service metadata documents repository and indexed in an inverted file. B. Political Divisions Service This element manages everything concerned with political divisions. Specifically, this service extracts political division candidates from every service metadata document stored in the repository and it sends requests to place names databases (Geonames [16] in our prototype) on the Internet for validating the division candidate and getting the hierarchy of its political divisions. In addition, the political divisions service is responsible for getting the geometry of every political division validated using OGC Services, specifically through gazetteers via WFS. The result of these procedures is the generation of a political divisions spatio-textual index where each division has related OGC services which provide information about such political divisions. In the following sections we describe this element. C. Tile Cache Gazetteer Service This third element in the crawler tier is in charge of obtaining as many non-political place names as possible through the WFS services found by the OGC Services Crawler. In this way we obtain a richer ontology with political divisions (as was shown in B) and non-political place names. In the following sections we also specify this element. (3) The search engine is the third tier of this architecture and it is able to work with the crawler tier switched off. The search engine is responsible for retrieving customized responses to the user’s queries. It has an index structure, previously created with the Crawler tier shown above, for resolving different scenarios in accordance with the user query type. The choice of which of these indexes has to be used in every scenario is made by the spatial and non-spatial IR module. This module is also responsible for adding semantic awareness to the queries by consulting the knowledge base. This knowledge consists of a set of taxonomies that represent the semantic relationships of meronymy and hyponymy between concepts of reality. Every taxonomy is built with the terms belonging to the same knowledge domain. In this study, these domains correspond to the categories defined in the Spanish Metadata Core [17] for describing geospatial resources. Some examples of knowledge domains are: hydrography, health, society and transport. It is not the aim of this paper to describe this search engine tier.

module will take both scenarios by default, and the final response will depend on which scenario has retrieved suitable results. If both of them have retrieved results, then the user has to make the decision of which is the scenario desired. This situation may arise with queries such as “river X”, where the system can interpret the user as being interested in a river called “X” but also the user could be interested in a map showing rivers crossing a village called “X”.

IV. CRAWLER TIER This section is focused on detailing index generation and index updating performed by the crawler tier. It is the main contribution of this paper. As has been mentioned above, the OGC Services Crawler module is in charge of finding these services published on the Internet. After validating every service found (by making a getCapabilities request), the XML metadata document response is stored in a repository. Then, at this moment a trigger is fired for updating the indexes.

Fig. 1. System Architecture

The main scenarios contemplated are: (1) Searches for specific divisions such as “London”, “California”, “Germany”, “River Thames” or “Lake Victoria”. The Spatial and non-spatial IR module obtains valid results using both the political divisions index and the non-political place names index matching the query. Every result in this scenario (1) is composed of a list with records that will have the name, the type (country, province, river, etc.) and the geometry. (2) Searches for concepts in a requested area such as “water bodies in the Netherlands”. In this case, what the user needs is a map that can be obtained through whatever indexed OGC Service which provides such information. In this case, the Internet tier will be necessary to analyze the query. For retrieving the custom map, the spatial and non-spatial IR module has to know if any of the words in the user query belong to any knowledge domain, so it is necessary to have the collaboration of a knowledge domain matching service that accesses and queries every knowledge domain taxonomy in the knowledge base to find out if any keyword in the user query belongs to any taxonomy. For the example “water bodies in the Netherlands”, the knowledge domain matching service will answer that “water bodies” belongs to the “hydrography” taxonomy, and therefore, all its hyponyms such as “river”, “lakes” or “ocean” will expand the query user. Note that the operator “in” is not considered as a spatial operator (preposition) and it will be discarded by the Internet tier. Our system does not interpret spatial relationships (in, near, intersect, contain, etc.). After taking a look at possible scenarios (1) and (2), the question arises of which scenario is suitable for the user query. To simplify the solution, the spatial and non-spatial IR

105

In the updating process, the next task is to find political divisions candidates in the XML metadata document provided by the getCapabilities request. These metadata documents give information related to the organization which provides such information, the layers provided and the service type (WMS, WFS, etc.). The task of finding political divisions candidates will be performed by the Political Divisions Service module parsing the XML tags. For example, in the WMS service the following tags will be parsed: “Abstract”, “ContactAddress” children and “Keyword”. After that, Geonames will be used as a validator of political division candidates and it will also work as a hierarchy provider. For instance, the word “Madrid” found in a metadata document (in one of the tags listed above) will be valid according to Geonames, and its hierarchy is Madrid is an urban area in the Autonomous Region of Madrid belonging to Spain. These steps are represented in Fig. 2. The Political Divisions Service module will proceed to update the political divisions and services index which has been implemented with an R-tree. We use R-tree because it is an efficient solution for spatial access methods and indexing multi-dimensional information in the same way as [10]. In this index, each node represents a political division and the relationships between nodes represent the geospatial relationship contain (we still do not consider other spatial relationship in our project). For instance, in Fig. 2, “Spain” contains “Autonomous Region of Madrid”. It is important to highlight that this index structure is seen as an ontology where the concepts are related with geographical information sources (layers belong to OGC services). Note that our approach constructs an ad-hoc semantic index structure (our ontology) and it does not use a generic spatial ontology, such as, for example, [19]. However, each node

representing a political division has an associated list of maps (layers), provided by the OGC services that offer information about that area (in the same way as [19]). These layers are also obtained from the XML metadata service document in which the political division name was found. It can be seen that the node “Madrid” in Fig.2 has its layer list associated to it.

Fig. 2. Political divisions and services index updating process managed by Political Divisions Service

Fig. 2 represents all the performed tasks explained above to build a political divisions and services index and an R-tree initial structure. As is shown in Fig. 2, all new nodes indexed (i.e “Madrid”) have no related geometry, yet (¿Geometry?). However, this geometry is mandatory to set up the R-tree index. For this reason, we have to find the geometry related to each node. Geonames provides the geometry of a given political division but this geometry is not enough since it is a point geometry and we are interested in a polygon geometry. The easiest way to get this polygon is by using any worldwide political cartography; however, in this approach we do not use previously stored cartography information, unlike [10]. Instead, our approach uses the OGC Gazetteer services (WFSG), whose metadata documents were already stored in the repository because such WFS-G services were previously found by the OGC Crawler service. WFS-G services have an operation to retrieve all information about a specific place name given, including the polygon geometry required. Thus, R-tree nodes with no related geometry can be fixed up and integrated. One of the most important aims of our approach is to provide semantic awareness to user queries. For instance, there may exist an OGC service found by the OGC services crawler previously offering a layer whose title is Hydrography with the term Spain (validated by Geonames as a political division) appearing in the service metadata document. Then, the political divisions and service index shown in Fig. 2 is updated by adding a layer named Hydrography into a related layer list of node “Spain” (this node already existed). In our approach, this hypothetical OGC service satisfies queries such as rivers in Madrid because Spain contains Madrid whereas river is a hyponym of Hydrography. We use the Knowledge Domain Matching Service module to provide this important semantic

106

awareness mentioned. As was shown in Fig. 1, a knowledge base in the search engine tier is available. This knowledge base has a set of taxonomies which describe several knowledge domains such as hydrography, transport or environment (one taxonomy per knowledge domain) in the same way as [13]. The Knowledge Domain Matching Service input is a concept composed of one or more words and the output is a list of hyponyms of such a concept. For instance, if the concept given is water bodies, the output will be a list with concepts such as river, lake or ocean. The same procedure is used to upgrade layer names associated with political divisions and services index nodes. If there is a layer named “water bodies 1:25000”, the Knowledge Domain Matching Service would replace its name with “water bodies, river, lake, ocean 1:25000”. Fig. 3 shows the final structure of the political divisions and services index. As is shown is Fig. 3, each layer in the layer list of a given political division node (i.e “Madrid”) has a related expanded name, also known as alias (i.e “Water bodies, river, ocean, lake 1:25000”). An id service, which is an integer number (i.e number “2” for layer “main motorways”), is also stored to relate each layer indexed with the service metadata document in the repository which has such an id. This is absolutely necessary in order to retrieve all the service information for handling future map user requests. Besides, the geometry related to each political division is a polygon (as we mentioned above). This index, shown in Fig.3, is enough to solve queries of political division location such as “Spain” or “Madrid” and also for returning maps with queries such as “motorways in Madrid”.

Fig. 3. Political divisions and services index- final state

Leaving behind the generation and updating of the political divisions and services index, the next step is to add a nonpolitical index in order to relate political concepts with nonpolitical concepts in our ontology. Non-political place names are necessary because one important scenario is represented by queries such as, for example, “Lake Victoria”. These kinds of queries cannot only be supported by the political divisions and services index. We need a new index, and the Tile Cache Gazetteer module generates this new index. Non-political

place names and their geometries are retrieved from WFS services. Specifically, the GetFeature WFS operation will be used. This operation returns features with fields such as place name (political and non-political) and its geometry in GML format. After a WFS service found by the OGC Crawler Service is validated, our approach continues with the following steps: (1) Obtaining the bounding box of every layer in the WFS service from the metadata document in the repository. (2) Splitting the bounding box into equally-sized pieces (tiles). This is advisable because a GetFeature XML response requested in a zone which is too wide could be very expensive, so it is better to reduce those requested zones. There is an option of making tiles in different scales in such a way that the lower the scale is, the greater the number of tiles generated will be, but this option is still being studied. (3) Making one GetFeature request per tile generated. (4) Adding a new record in the non-political place names index per feature retrieved. (5) Calculating the political divisions where the added nonpolitical place name is located. This task is performed with an intersection operation between the non-political place name geometry and the geometry stored in each of the political divisions and services index nodes.

Fig. 5. Political and non-political indexes - final structure

In Fig.5 a new element stands out, namely a hash table that provides direct access to every node in the R-tree through the political division name. This hash table has been inserted to avoid a complete search throughout the R-tree when the place name is the argument given in the query.

V. USER QUERIES In this section the process performed by the search engine tier to answer user queries is detailed. The spatial and non-spatial IR module manages the user queries as shown in Fig.1. The following examples are run with the index content and structure shown in Fig. 5. In addition, these examples correspond to the scenarios defined in section III:

Fig. 4. Political and non-political indexes relationship

Fig. 4 shows both political and non-political indexes and how they relate to each other after our approach applies the process. Focusing on the non-political place names index, each record (i.e “Manzanares river”) has a related geometry which was obtained through the WFS service; it also has a related list with the political divisions which intersect with it and, finally, an id service which is an integer number (i.e number “2” for place name “A3 motorway”) to relate each place name with the service metadata document in the repository which has such an id. As is shown, it is possible to obtain all non-political divisions located in a given political division and vice versa. This non-political index can be used for user queries requesting, for example, “A3 Motorway” or “Manzanares River” locations. It is very useful for users to be able to find, for instance, regions and provinces that the Manzanares river crosses. This kind of complex spatial analysis is being researched. The whole final index structure is shown in Fig. 5.

107

(1) User query “Madrid”: This term has several records in the hash table (“Autonomous Region of Madrid”, “Province of Madrid” and “Madrid”), so this term is a valid political division. No terms are found in the non-political index with “Madrid”. Thus, the system response is a sorted list joining the results provided by the political index (3 results) and the non-political index (no results). Political divisions are listed before non-political ones and the lower political divisions will be listed first. So, the response is {(Madrid, Urban Area, geometry), (Province of Madrid, Province, geometry), (Autonomous Region of Madrid, Region, geometry)}. The political division type (Urban area, Province, Region, etc.) is provided by the corresponding level in the R-tree in which a given node is located. Each result, Province of Madrid for example, has an associated layer or layers (and their related service) with spatial information for this location. As an added value, the type of each layer retrieved is indicated (raster if it comes from a WMS service, or vector if it comes from a WFS service). (2) User query “Manzanares”: This term has no records in the hash table, so it is not a valid political division. However, it has one result in the non-political index (“Manzanares river”). The response is {(Manzanares river, non-political,

geometry)}. Besides geometry, each result has an associated WFS layer and its related service, which shows the nonpolitical place queried. (3) User query “Roads in Madrid”: After preprocessing, the user query is just “Roads Madrid”. “Roads” is neither a political division nor a non-political division. The Knowledge Domain Matching Service is used to check if “Roads” belongs to any taxonomy. The result is that “Roads” belongs to transport taxonomy and the query is expanded to “Roads, motorways, conventional roads, Madrid”. Otherwise, “Madrid” is a valid division. So, our approach interprets that it is a query with the form concept of reality (roads) in a zone (Madrid). In this case, the search engine looks for layers in the Madrid R-tree node matching the expanded query. The result is (Main motorways, Motorway, layer). Furthermore, after expanding the query, a non-political division located in Madrid has been found (A3 Motorway). So, the final answer is {(Main motorways, Motorway, layer), (A3 Motorway, Motorway, geometry)}. Besides geometry, each result has a related layer and service in the same way as for the queries above.

VI. CONCLUSIONS AND FUTURE WORK We have presented an approach to integrate spatial resources provided by OGC services in an ontology. This resulting ontology (ad-hoc construction) relates political divisions, nonpolitical place names and layers offered by OGC services. Furthermore, semantic awareness has been added to the solution, thus providing a powerful tool for the search engine. We have seen how all the index structure is automatically generated by the crawler modules. Finally, we have described several scenarios of use and how modules in the architecture collaborate to find suitable responses to user queries. This approach makes it easier to discover geographical information offered by SDIs. Furthermore, other GIS applications can benefit from this spatial search engine by showing suitable geographic resources to meet individual needs. Future work foresees the carrying-out of a performance study in order to validate the quality of the ontology (corpus). Furthermore, we are working on including political division candidates which are not explicitly in the service metadata documents. Finally, the query results order should be made to depend on parameters such as QoS, popularity and the content related to the OGC service which provides the layers retrieved. ACKNOWLEDGMENT This research is funded with the Subprogramme “Torres Quevedo” (MICINN-PTQ) from the Ministry of Science and Innovation of Spain (Ref. PTQ-09-02-02242).

108

REFERENCES [1]

J. Berners-Lee, J. Hendler and O. Lassila. The Semantic Web, Scientific American, vol 184, no. 5, pp. 34-43, 2001. [2] J. Egenhofer. Toward the Semantic Geospatial Web. ACM-GIS 2002. 10th ACM International Symposium on Advances in Geographic Information Systems. McLean (USA). 2002. [3] O. Boucelma, M. Essid and Z. Lacroix. A WFS-Based Mediation System for GIS Interoperability. ACM-GIS 2002. 10th ACM International Symposium on Advances in Geographic Information Systems. McLean (USA). 2002 [4] A. Gupta, R. Marciano, I. Zaslavsky and C. Baru. Integrating GIS and Imagenery through XML based information Mediation. Integrated Spatial Databases: DigitalImages and GIS. Lecture Notesin Computer Science. Vol1737. Pp. 211-234. Springer-Verlag. 1999 [5] J. Córcoles and P. González. Querying Spatial Resources. An Approach to the Semantic Geospatial Web. CAiSE'03 workshop "Web Services, eBusiness, and the Semantic Web (WES)”. To Appear in Lecture Notes in Computer Science (LNCS) by Springer-Verlag. 2003. [6] J. Córcoles, P. González and V. López-Jaquero. Integration of Spatial XML Documents with RDF. International Conference on Web Engineering. Spain. To Appear in Lecture Notes in Computer Science (LNCS) by Springer-Verlag. 2003. [7] C. Jones, A. I. Abdelmoty, D. Finch, G. Fu, and S. Vaid. The SPIRIT Spatial Search Engine: Architecture, Ontologies and Spatial Indexing. Proceedings of the 3rd International Conference on Geographic Information Science, pages 125–139. 2004. [8] M. D. Lieberman, H. Samet, J. Sankaranarayanan and J. Sperling. Steward. Proceedings of the 15th annual ACM international symposium on Advances in geographic information systems - GIS '07, New York. ACM Press. 2007. [9] A.I Abdelmoty, P. Smart and C.B Jones. Building place ontologies for the semantic web: Proceedings of the 4th ACM workshop on Geographical IR - GIR '07. New York. ACM Press. 2007. [10] M.R Luaces, J.R Paramá, O. Pedreira and D. Seco. Retrieving Documents with Geographic References Using a Spatial Index Structure Based on Ontologies. International Conference on Scientific and Statistical Database Management, SSDBM 2008, pages 384-400. 2008. [11] N. Chen, J. Gong, and Z. Chen. A High Precision OGC Web Map Service Retrieval Based on Capability Aware Spatial Search Engine. To Appear in Lecture Notes in Computer Science ( LNCS), vol. 4683, pages 558–567. Springer Berlin, Heidelberg. 2007. [12] G. Lan and Q. Huan. Ontology-based Method for Geospatial Web Services Discovery. Proceedings on Intelligent Systems and Knowledge Engineering (ISKE2007). 2007. [13] H. Hochmair. Ontology Matching for Spatial Data Retrieval from Internet Portals. Proceedings of the First International Geospatial Semantics Conference. Berlin, Springer Lecture Notes in Computer Science (LNSC) No 3799: 166–82. 2005. [14] OGC 02-069, OpenGIS implementation Specification: Web Map Service, http://portal.opengeospatial.org/files/?artifact_id=4756 [15] OGC 02-058: OpenGIS implementation Specification: Web Feature Service, http://portal.opengeospatial.org/files/?artifact_id=8339 [16] Geonames: Gazetteer. http://www.geonames.org. [17] OGC: OpenGIS implementation Specification: Web Gazetteer Service, https://portal.opengeospatial.org/files/?artifact_id=7175. [18] GSDI: Global spatial data infraestructure association. Retrieved July 2010 from http://www.gsdi.org. [19] J. E. Córcoles and P- González. Integrating GML resources and other web resources. 1st International Workshop on Geographic Information Management (GIM'04) in conjuntion with DEXA'04 published by IEEE Computer Society Press. Zaragoza. Spain. 2004 [20] Open Geospatial Consortium (OGC): http://www.opengeospatial.org. [21] Alia I. Abdelmoty , Philip D. Smart , Christopher B. Jones , Gaihua Fu , David Finch, A critical evaluation of ontology languages for geographic information retrieval on the Internet, Journal of Visual Languages and Computing, v.16 n.4, p.331-358, August, 2005.


Prototype of Semantic Search Engine Using Ontology Ahmad Maziz Esa, Shakirah Mohd Taib, Nguyen Thi Hong Computer Information Sciences Universiti Teknologi Petronas Tronoh, Perak [email protected], [email protected], [email protected] Abstract—In this paper we discuss the fundamental problem of information retrieval on the Web. Information on the Web is not semantically categorized and stored. This research focuses on applying semantic capabilities using ontology on search engine. By using ontology, search engine can search keywords that are conceptually linked instead of just similarity of the words used. This paper also provides in depth description of the architecture design of our proposed modified search engine. This paper describes how the mechanism is designed so that the search engine can extract information stored based on the ontology and present a semantically linked search results. The benefits and future improvements are also discussed.

directory provider. It was Google who among the first that implement automate indexing and crawling mechanism which enables the search engine to automatically crawl Web pages and indexed the retrieved Web pages for users to search [11]. Google uses page rank by keeping track the number of incoming links and links linked to other pages. The more links linked to a page the more credible the page is, thus will be ranked higher than the other. All these were being computed using mathematical algorithms by calculating the term frequency and inverted term frequency. Data crawled and collected were stored in an inverted database which enables the search engine to locate which terms were stored in which document or links. There is no doubt that the collection of information on the Web is increasing. As for now, with the current search engine which utilizes on mathematical algorithms will be able to cope. As the collection of the information becomes larger, it will dilute the accuracy of conventional search engine making it less accurate and less precise. The dilution of the result accuracy will be further aggravated as the collection of information grows rapidly. This work aims to tackle the problem from a different angle. Instead of trying to preserve the accuracy of search engine by relying on machines speed and processing power to enable the usage of more sophisticated mathematical algorithms, this work will explore semantic utilization in search engine. By implementing semantic mechanism in search engine, it will enable information to be related to each other conceptually. This will give the information indexed with the semantic value which can improve information retrieval. Major contributions of this work are basically the analysis of semantic search engine and the development of semantic search engine prototype which enables user to have more accurate search semantically. Section 2 describes related works done. Section 3 describes the methodology used to analyze and develop a semantic search engine. Section 4 describes the architecture and algorithm used in order to provide semantic mechanism for the search engine. Section 5 concludes the paper and finally section 6 describes future improvements to this work.

Keywords-component; search engine, semantic, information retrieval,ontology.

INTRODUCTION The Web at its infancy was a static page which allows users to open and read the contents of the Web pages. There was only a one-way interaction between the users and the Web. As the technology advances, Web-enabled devices were getting cheaper and more ubiquitous. More and more people are able to access the Web and utilize the wealth of information in it. This triggered a paradigm shift in Web usage and the way people interact with the Web. Experts and laymen coined this shifting in Web interaction as Web 2.0. A Web 2.0 site enables users to interact with the Web more interactively with still and moving graphics as well as sound [1]. Users were also able to publish their contents for the consumption of other users. It gave way to the birth of Web 2.0 technologies such as Friendster [2], Youtube [3], Blogger [4] and Facebook [5]. It was an age of ―content by the users for the users‖. Content creation were not limited to just an organization but also to anyone who has access to the Internet. As envisioned by Tim Berners Lee in his book titled Weaving the Web [6], the Web will implement semantic properties in its collection of Web pages which will understand the words and terms human used. The large amount of information on the Web can be retrieved using a search engine. Since Web 1.0, many search engines were developed and been commercialized. These search engines such as Google[7], AskJeeves[8], Yahoo![9], and Lycos[10] were among the search engines that were dominating at its time. Search engines help users by indexing all the information on the Web and make it easy and quickly retrievable for the users. Early search engines were not a search engine at all. Instead it was a directory which contained indexed information which were indexed manually by the

978-1-4244-9191-9/10/$26.00 ©2010 IEEE

RELATED WORKS Currently, a general purpose ―semantic‖ search engine had been developed. The search engine can be accessed at www.hakia.com. However, most of the mechanism used in the search engine were patented and focused on commercial use. As quoted by Tim Berners Lee [6], ―I mention patents in passing, but they are a great stumbling block for Web

109

development‖. All the technologies used in Hakia[12] were patented and therefore are trade secret. This prevent academic circle to study intricate workings of the search engine for future improvement and other applications. Many search engines have been developed throughout the years. One of the most dominant was Google[7]. Many the components work together to produce search results. In the architecture, Google implemented Page Rank algorithm to identify the relevancy of the result. Pagerank algorithm will be explained further in the next section. The crawlers work 24/7 traversing all the hyperlinks and downloading Web content into storage. All the contents are parsed and indexed and stored into another storage area. The index is then inverted to allow each single term to be related to many words. The PageRank algorithm ranks the Web pages based on citation principle. The more links referred to a particular link, the higher the point it will have. The weight of Web pages will also be taken into account. If a Web page with a high weight is a reference to a Web page, it will have higher points. The higher point will result in PageRank to rank higher in the result. The PageRank calculates the links from all pages equally and normalizing it. The basic formula is BRIN [3]: PR(A) = (1-d) + d (PR(T1)/C(T1) + …… + PR(Tn)/C(Tn))

It requires Tomcat as Servlet container. The details of Nutch’s skeleton is discussed in part IV. Nutch leverages on distributed computing to process large data sets [15]. The distribution file system it’s using is Hadoop [16] which is also used by Yahoo! for its search engine system. Hadoop uses a programming model call MapReduce which was developed by Google [7]. In the model, it uses a set of as computation inputs. This input is used by map function to parse the task and generate intermediate keys. These intermediate keys will be the input for reduce function and merges together similar key and produce an output. Nutch parse task for fetching, crawling and indexing into this set of key and value which will be replicated on various slave machines and will be computed [15]. The result will then be merged together in designated location for usage by the searcher. Liyi Zhang [17] had conducted a research of using ontology to improve search accuracy. The retrieval system is an E-Commerce product retrieval system which uses ontology-based adoption Vector Space Model. It modified existing vector space model to treat documents as a collection of concepts instead of documents as collection of keywords. To determine the similarity between the documents and user query, it uses weights that are calculated using tf-idf(term frequency, in this case concept frequency and inverted document frequency) scheme. According to Liyi Zhang [ibid], the system conduct parallel searches using OA-VSM and SPARQL information retrieval. Both of the result will be matched and ranked and the best result will be presented to the user. Our project focuses on development of search engine model extension named Zenith. This extension is a plug-in for Nutch [18] which enables it to function as a semantic search engine. By integrating Zenith and Nutch, they work together as a hybrid semantic search engine which can be used as a proof of concept for our research.

Where PR(A) is the probability of Web Site A, which contain T1 pages… to Tn. T1 - Tn are pages linking to page A. PR(T1) is the PageRank value for page T1. D is the damping factor that can be set to 0 to 1. C(A) defines number of links going out of Web site A. Nutch is an open source search engine which was developed by Doug Cutting[13]. Nutch is an extension of Lucene [14] which is an open source information retrieval system. Most of the Lucene libraries were used in Nutch. Most of the package in the figure provides Nutch functionality such as indexing and searching capabilities. According to Cutting [13], Nutch consist of two main components: a) Crawler 

Webdb



Fetcher



Indexer



Segments

METHODOLOGY Zenith development uses a combination of reusable prototyping and component-based development as shown in Figure 1. The development begins by conducting literature review. Components that can be reused in this project are also identified. This process is called domain engineering. Domain engineering is a process of identifying the software components that is applicable for Zenith’s development [19]. Each Zenith’s function is compartmentalized into components. In the component sub-phase, reusable prototyping model is implemented. In general, the whole system is basically a reusable prototyping. This methodology is most suited for Zenith development because of a few factors. Zenith architecture is highly modular. Components from other past projects can be reused in Zenith’s development. As mention above, Zenith development is highly unpredictable. This methodology facilitates unpredictability of Zenith development. For instance, this methodology allows developer to experiment

b) Searcher Webdb is a persistent database that tracks page, relevant link last crawled date and other facts. In addition, Webdb stores image of Web graph. Fetcher on the other hand, is what made crawler it is. Fetcher basically, crawls from one Web site to the other and fetch the content back to the system. The indexer uses the content fetched by fetcher to generate an inverted index. The inverted index is then divided into segments which than can be used by searcher to display query results. Searcher components provide the interface for users to conduct search.

110

with components and methods and test selected components as proof of concept. The way this methodology flow allows developer to go back to previous phase to conduct modifications. In addition, the risk of development progress being hindered by developer mental block will be reduced as developer can shift the development efforts to other components.

A. Back-end Back end is where the process of getting and storing information gathered from the Web. Majority of the core functions and search engine capabilities depended on how the backend is designed. Backend has Web crawler, URL Server, Indexer and the storage. Web Crawler Web crawler is a script which is executed to retrieve Web pages based on the URL list stored in the URL server. To make Web crawling more effective, Web crawlers must be implemented in a way where many crawlers can simultaneously crawl multiple Web from the URL server. Threading implementation is required to enable concurrent processing. Another important functionality is the crawler has to be able to understand robot exclusion protocol. Webmaster that wants their site to be excluded from being crawl by crawlers can include robot exclusion protocol file. This file is a text file format which outlines which part of the site tree that is not accessible for the crawlers. Webmaster also can specify ways to guide the crawlers to crawl their Web site. Finally all the parsed data are store in a database of URL Server URL server acts as the storage for URL links. A list of URLs of commonly visited sites is manually stored and becomes the starting point for the crawlers. New URLs found by the crawler will be stored in the server.

Figure 1. Zenith Methodology

Aside from adapting to the development requirements of the system, this methodology will increase the system maintainability, and scalability. A highly scalable system will be able to cater large number of users in accordance to the system resources it can use. Maintainability is important to keep Zenith relevant in the future. When it is maintainable and scalable, the system can easily be enhanced for reliability. Fault tolerant capabilities can be implemented by making each component redundant. The introduction of component redundancy increases the system performance through loadbalancing. All in all, this methodology is designed and modified to specially suit the nature of Zenith development. Even though this model tries to capture as much development activities as possible is does not capture all.

Indexer An indexer functions by indexing the parsed data into its according type. Indexer will ―organize‖ the data into categories. A document found by crawler will be parsed and indexed with a unique id, its data type, file and content. Indexer must be able to parse HTML, PDF, Words and other documents found from the crawling activities. The data from the parsed HTML will be extracted and stored into a storage area such as database or custom data storage. Most implementations of search engine will compress all the data in order to maximize storage space.

ANALYSIS OF GENERIC SEARCH ENGINE SKELETON Generic search engine skeleton describes the back bone of search engine. It consists of features and functionalities necessary for it to be identified and function as search engine. Generic Search Engine Skeleton had been derived from educational and experienced conjecture. In this research, Nutch [18] will be used as conventional search engine prototype. This prototype will enable better understanding of the mechanism and the nature of search engine. From the prototype, a semantic search engine design will be derived. This design is drafted in the proposed design section. A search engine consists of few major key components. These components are divided into two sections which are the front end and the back end. The main components were analyzed based on the architecture proposed by Brin [20] and analyzed by Manjula [21]. Nutch[18] is designed based on the skeleton that has a Back end and a Front end sections. We use this skeleton as the basis of the proposed design for semantic search engine. Few components are added and modifications of the skeleton components are made in order to implement semantic capabilities.

B. Front End Front end has only one main component which is the searcher. This component acts as an intermediary between the user and the system. The component provides users the interface to obtain user search keyword. The keyword is then search in the reversed index which then will point to the links to the site. Multi-threading capabilities is required in the searcher as many users will conduct search simultaneously. OVERALL ARCHITECTURE Zenith is an expansion for Nutch[18] that enables it to be a hybrid semantic search engine. The original design of data flow will be intercepted and modified before being rechanneled to the indexer. Figure 2 shows the overall architecture of Zenith model.

111

1.

Get data from Nutch

2.

Extract Data from ontology using Jena

3.

Run inference engine on the extracted ontology

4.

Compare Subjects from document (Index Predicate if Yes)

5.

Compare Predicate from document (Index Subject if Yes) Iterate if there is still more class

6.

Re-channel data to flow

Figure 2. Nutch with Zenith Expansion Architecture

This is enabled by the design of Nutch which implements plug-ins capabilities. Zenith expansion acts as Nutch’s plug-in which will be called when the data is being indexed. Developing Nutch plugin involves extending IndexerFilter extension interface provided by Nutch. Semantic Indexer implements IndexFilter interface. SemanticIndexer.java skeleton is as shown in Figure 3:

Crudely the essence of the mechanism to implement the semantic capability is by utilizing ontology. The mechanism lies on how statements are extracted and manipulated to find semantic relation. Using external framework called Jena[22], ontology information is extracted in the form of: “Subject Relationship Predicate” Subject represents classes or entity that the statement is describing. It could be bank name, payment method or etc. Predicate represents object or entity that the statement is describing. Relationship describes how subjects and predicates are related. Semantic indexer will use the subject item as key term when searching documents crawled by crawler. Once found, it will then index the predicate in the index. Once all subject terms are completed it will repeat the process using predicate and index the subject instead. Currently there are two types of relationship, positive and negative relationship. Positive depicts related relationship while negative depicts unrelated relationship.

Figure 3. Semantic Indexer Skeleton

The main Semantic Indexer will instantiate Jena framework and Xerces and will then extract data from ontology and ontoIndex accordingly. These data is stored in memory for manipulation subsequently. As mentioned above, Semantic Indexer performs its tasks by calling other components which is explained in the next section.

ZENITH EXPANSION ARCHITECTURE In Zenith expansion there are few components that work together to give Nutch[18] the ability the semantic ability. The components in Zenith expansion are semantic indexer, Jena framework, Ontology, Xerces and OntoIndex .

D. Jena Framework Jena framework is an open source semantic Web framework [22]. It enables the Semantic Indexer to extract data from .owl file (an ontology) for query and manipulation. It also provide semantic indexer a reasoning engine which will infer the ontology contained in the ontology by adding rules to it. This established the logical rules based on the relationships in the ontology, hence assist the search engine to better define the semantic relationship between the concepts.

C. Semantic Indexer All the core functionalities and the algorithm that orchestrate the process getting the data from the indexer, extract information from ontology using Jena [22] and incorporate semantic value in the data. The data then is channel back to the indexer to be passed to the searcher which that will be displayed in the search result. The algorithm is as shown in Figure 3.

112

E. Ontology Ontology is the component where the semantic is derived from. It acts as a ―brain‖ or central location where the source of semantic or ―knowledge‖ is from [23]. The wider or broader the subject scope of the ontology the more search engine can derive terms to index semantically.

The ontology is built based on Methontology (Fernandez)[25?]. The methodology includes steps and activities carried out in several cycles. Each cycle includes three main types of activities: management, technical activities, and support activities. Once ontology is drafted, it is tested using the search engine. Then the ontology is modified accordingly to best support the search engine’s efficiency. Management work is about planning of the objectives and its users. Technical activities include specifying the scope and granularity of the ontology. The granularity level means how detail the ontology is. In this case it has two levels of subclasses. The scope of the ontology is identified in Business To Consumer E-commerce area of knowledge. Then all the knowledge of the scope was then conceptualized. Key terms and concepts are identified and linked by their relationships (called property). Then the ontology is constructed using software names Protégé. Besides that, support activities are carried out. Support activities are mainly knowledge acquisition of business to consumer e-commerce, and ontology fundamentals. Verifying the knowledge gained with domain expert is also an essential part of support activities.

F. Xerces Xerces is an open source library which enables the Semantic Indexer to extract data from .xml files [24]. The library is used to extract data from ontoIndex which is used in retrieving semantic information from the ontology before indexing it in the index. G. OntoIndex The OntoIndex is an index file act as a point of reference for Semantic Indexer to refer to when iterating through the ontology data. The ontology index stores information in tag with name and value. The value tag contains multiple value that is separated by ―|‖ without the quote. The ontoIndex contains information such as class name and relationship used.

I.

Results Two processes of indexing were conducted with semantic indexer enable and disabled. Table 1 shows the result of testing with conventional model and semantic enabled model. The result shown is a comparison of data indexed by search engine.

PERFORMANCE EVALUATION H. Methodology This section highlights the test conducted to see the effectiveness of the semantic mechanism implemented in Nutch. A test dataset was developed using html files with hyperlinks. The search engine and the test data site are on the same machine served by tomcat and apache server as in Figure 4. We developed ontology of e-commerce as our subject scope for the testing.

Table 1. Difference in amount of data indexed when enabled and disabled Parameters Disk Space (bytes) No. Terms No. Documents

The hierarchy of the test data is shown in Figure 5.

Ecommerce index

With Semantic Capability 224404 4654 9

From the result in Table 1, it shows there is an increase in disk space by 0.06% and number of terms by 0.03%. This shows data being index is more compared when the semantic capability is disabled.

Figure 4. Test Architecture

Ebusiness

Conventional 217235 4525 9

Table 2 shows the search result comparison using sample search keywords.

crm

Table 2. Search Results Comparison

businesstoconsumer

PaidContent

Conventional Semantic Indexer (Zenith) Keyword = ―b2c_ecommerce‖ 0 6 Keyword = ―book‖ 1 6 Keyword = ―event_ticket‖ 0 7

Googlebuys AtomServers Nuclearphysics Figure 5. Hierarchies of Test Data

As a model, the ontology used for the search engine does not reflect the real world data and naming convention. Keyword used can be replaced with bank names, book or movie title to

113

reflect real world entity and thus real world semantic relationships. Even though it does not reflect real world, as a proof of concept it is adequate to say that the semantic search engine is possible. Based on Table 2, semantic search engine are capable of returning information that conventional search engine cant because conventional search engine returned results that contains instances of the keyword while semantic search engine returned results that contained instance of the keyword and results that related semantically.

CONCLUSION In conclusion, the research has met the objectives set. From the result of experiments and analysis, it is found that Nutch with Zenith expansion can become a full-fledge semantic engine by modifying its architecture. As more technologies being developed such tools and software (Wordnet, Hadoop, etc), Nutch can utilize it and further improve its semantic capabilities. As a conclusion, the analysis and discussion above remained non-conclusive. Further research, studies and testing might result in changes of the architecture and design of Zenith.

J. Issues Limited Amount of data Due to limited resources, only small case test with limited amount of data can be conducted. This could not exhibit the full potential of Zenith and the full extent of the problem in Zenith when handling large amount of data.

REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11]

Buggy Scoring System Scoring system helps search engine to organize search result according to ranks which is based on the importance of the documents or site indexed. Due to the implementation of Zenith, it had disrupted the scoring system in Nutch. Although the result will displayed site or documents that contains instance of keyword or semantically related to the keyword, it would not displayed according to rank of importance. To resolve this, the scoring system may be modified.

[12]

FUTURE IMPROVEMENT

[13]

K. Artificial Intelligence In the realm of Artificial Intelligent there are technologies such as fuzzy logic, neural networks and genetic algorithm. Fuzzy logic can be use in Nutch searcher. Most words in English are ambiguous with many meaning depending on context. Together with fuzzy logic, Genetic algorithm can be used to determine the accuracy of the keyword and the terms stored in the documents in Nutch index. It also can be used together with Semantic Indexer to relate the concepts in ontology more accurately. Lastly fuzzylogic can equip Nutch with the capability to learn new concept from data crawled from the Web.

[14] [15] [16]

[17]

[18] [19] [20]

L.

Wordnet Wordnet is a lexical database for English language. It contains synonyms of words and synset which can be used to identify the connection of one word with others and type of words such as adjective or verb [26]. It can be used to enhance Zenith to not just search for semantically related but also related based on lexical rules.

[21]

[22] [23]

M. Distributed Computing Nutch is equipped with distributed computing facilities using Hadoop [16]. The Semantic Indexer can be modified to leverage on distributed computing to process the ontology inference process and term comparison processes.

[24] [25] [26]

114

S. Murugesan,‖Understanding Web 2.0‖, IT Pro, IEEE Computer Society, July 2007, pg. 34 – 41. Friendster, Available: http://www.friendster.com Blogger, Available: https://www.blogger.com Youtube, Available: http://www.youtube.com Facebook, Available: http://www.facebook.com T. Berners Lee, ―Weaving The Web‖ 2000, pp. 191-215. Google Search Engine, Available: http://www.google.com Askjeeve Search Engine, Available: http://www.ask.com Yahoo Search Engine, Available: http://www.yahoo.com Lycos Search Engine, Available: http://www.lycos.com/ S. Asadi and H. R. Jamali, "Shifts in search engine development: A review of past, present and future trends in research on search engines". Webology, 1(2), 2004, Article 6. Hakia - Ontological semantic and natural language processing (NLP) based search engine., Availble: http://www.hakia.com D. Cutting and M. Cafarella, ―Building Nutch: Open Source‖, Focus Search, April 2004, pg. 54 - 60 Lucene – Text search engine library in Java, Available: http://lucene.apache.org J. Dean and S. Ghemawat, ―MapReduce: Simplified Data Processing on Large Clusters‖, Whitepaper, Google, Inc. Shvachko, K. , Kuang, H. , Radia, S. ; Chansler, R, The Hadoop Distributed File System, IEEE 26th Symposium on Mass Storage Systems and Technologies (MSST), 2010 L. Zhang, M. Zhu and W. Huang, ―A Framework for an Ontology-based E-commerce product information retrieval system‖, Journal of Computers, Vol 4, 2009, pg 436-443. Nutch - Open-source Web-search software, built on Lucene Java, Available: http://nutch.apache.org/ Nauman, J.D. and Jenkins, M. ―Prototyping: The New Paradigm for Systems Development‖, MIS Quarterly Vol. 6, Ed. 3, pg. 29-44 S. Brin, L. Page, ―Anatomy of a Large-Scale HypertextualWeb Search Engine,‖ Proc. 7th International World Wide Web Conference, 1998. D.Manjula and T. V. Geetha, ―Semantic Search Engine‖, Journal of Information and Knowledge Management, Vol. 3, No. 1 (2004) pp. 107117. Jena – A Semantic Web Framework for Java , Available: http://jena.sourceforge.net/ M. Uschold and M. Gruninger, “Ontologies and Semantics for Seamless Connectivity‖, ACM SIGMOD Record COLUMN: Special section on semantic integration, Volume 33 , Issue 4, 2004, Pages: 58 – 64. The Xerces Java Parser 1.4.4, Available: http://xerces.apache.org/xerces-j. A. Gomez, M.Fernandez, O.Corcho. ―Ontological Engineering‖, Spinger 2003, pg 3-5 Wordnet, A Lexical Database For English, Available: http://wordnet.princeton.edu/wordnet.