Hierarchy Database Lexicon Solution for Sentiments ...

A Database Solution for sentiments challenge

Hierarchy Database Lexicon Solution for Sentiments Challenge Doaa Mohey El-Din Information Systems Department Faculty of Computers and Information, CU Cairo, Egypt [email protected]

Abstract— analyzing online sentiments challenges become hot research area to improve the accuracy and ease to understand. There are theoretical and technical sentiment challenges. World knowledge interprets old knowledge and updatable information. Till now, a few research in this challenge, because of the hardness of it and no standard measurement for it. This challenge clearly appears in two issues: recent events or linguistics similarities. This paper presents a new lexicon which is a solution for handling world knowledge challenge. This lexicon relies on a hierarchy database model. This research presents a new relationship between a topic domain and world knowledge challenge. Sentiment analysis plays a vital role in business decisions. Our target involves this importance in a scientific domain to support the researchers. The experiment concentrates on linguistics similarities and knowledge information not updatable. Its results achieve nearly 70%. Keywords— Sentiment analysis; Reviews; Challenges; Implicit Negative; Explicit Negative;.

I. INTRODUCTION Analyzing online sentiments is a vital role in interpretation information into a score. It requires using the natural language processing, text analysis, and computational linguistics to define and extract information. Sentiment analysis [1] is widely applied to online reviews on several websites or social media. The goal of sentiment analysis is evaluating the sentiments respectively sentiment analysis classification. There are several challenges face the evaluation of sentiment analysis. Sentiment analysis challenges divide into two types [2]: Theoretical and technical. These challenges become obstacles in analyzing the accurate meaning of sentiments and detecting the suitable sentiment polarity [3]. A new research area is sentiment analysis challenge [2]. World knowledge [4, 5] refers to facts, information, recent information such as political elections [6], or linguistics similarities [7] such as "the actor likes a lion in this movie". The challenge of world knowledge splits into two types: updatable information and linguistics similarities. We suppose that text reviews is a particularly important modality for sensing affect because the bulk of human opinion textually based. It keeps track of the development of robust textual affect understanding meanings. In addition, improved textual meaning can enhance the accuracy of sentiment analysis evaluation including several modalities, like facts or speech expressions.

The proposed technique [8] improves a sentiment accuracy through handle several sentiment challenges. This paper introduces a solution to enhance a world knowledge accuracy. This research includes facts, knowledge and linguistics similarities. The proposed solution cares of the grammar of each review sentences. It combines between bag-of-words model and Part-of-Speech model. Our proposed lexicon relies on two parts: nouns and others. The others refer to any part of speech expect nouns such as verbs, adverbs, adjective, Etc. We can handle these parts of speech with using similarities and differences algorithms to achieve infinitive verbs. But the hardest issue in constructing lexicon is noun part. The noun speech has several types which have keywords, facts, other features. We discuss the close relationship and effect between the topic domain and world knowledge sentiment evaluation. The paper is structured as follows: Section 2 represents sentiment analysis. Section 3, the related work discussion of the world knowledge challenge. Section 4, the presentation of a new framework. In Section 5, outlines of the Experiment. Finally, Section 6 conclusion and future work. II. SENTIMENT ANALYSIS The purpose of this paper shows analyzing sentiments and evaluating them. The authors [9] presented a tool which judges the quality of text based on annotations on a scientific papers. Its methodology collective’s sentiment of annotations in two approaches. It counts all the annotation produces the documents and calculates total sentiment scores. Its problem declares in a relationship between annotations that is complex. The technique needs to have a big query knowledge base containing metadata. Our notion that values are not accurate enough and have some logical errors such as the value of “Good=0.875” has greater value than the value of “Best=0.75”. Nevertheless, we believe that collecting metadata and evaluating them could be useful to achieve to higher analysis quality. The researchers proposed a mining system for hotel reviews [10]. They introduced an evaluation system for online user’s reviews and comments to support quality controls into hotel management. It is capable of detecting and retrieving reviews on the web and deals with German reviews. It has multi topic domain and based on multi polarity classification; the system would recognize the neutral e.g., “don’t know” to “classify sentiment polarity that as neutral” and the multi-topic

A Database Solution for sentiments challenge

cases identified in their corpus. It is most weakness illustrate in not handling some cases in multi-topic segments. III.

WORLD KNOWLEDGE CHALLENGE

World knowledge often requires to be incorporated in the system for detecting sentiments. Consider the following examples: "He is Zewail in this research". Just finished Doctor Zhivago for the first time and all I can say is Russia sucks. The first sentence depicts a negative sentiment whereas the second one depicts a positive sentiment. But one has to know about Frankenstein and Doctor Zhivago to find out the sentiment. The main task in this approach is the construction of word lexicons that indicate positive class or negative class. The sentiment values of the words in the lexicon are determined prior to the sentiment analysis work. Lexicons can be created in different ways. It can be created by starting with some seed words and then using some linguistic heuristics to add more words to them, or starting with some seed words and adding to these seed words other words based on frequency in a text. SentiWordNet 3.0 is a publicly available lexical resource explicitly devised for supporting sentiment classification and opinion mining applications [11]. The paper [12] proposed to find subjective sentences using lexical resources where the authors hypothesize that subjective sentences will be more similar to opinion sentences than to factual sentences. As a measure of similarity between two sentences, they used different measures including shared words, phrases, and the WordNet. The research in [13] focuses on extracting top sentiment keywords which is based on Pointwise Mutual Information (PMI) measure. Until now, a few research in this challenge, because of the hardness of it and no standard measurement for it. We introduce a new solution for enhancing the accuracy of sentiments. IV. PROPOSED SOLUTION The proposed solution focuses on linguistics similarities, facts and knowledge information. These solutions presents a new lexicon which relies on a hierarchy database model. This research presents a new intimate relationship between a topic domain and world knowledge challenge. Our target involves this importance in a scientific domain to support the researchers. Hierarchal database lexicon constructs based on nouns. It supports to detect the sentiment polarity and interpretation score to ease to understand meaning accurately. In a hierarchical model [14, 15], data is organized into a tree-like structure, implying a single parent for each record. A sort field keeps sibling records in a particular order. Hierarchical structures are widely used in the early mainframe database management systems. This structure allows one one-to-many relationship between two types of data. This structure is very efficient to describe many relationships in the real world; recipes, table of contents, ordering of paragraphs/verses, any nested and sorted information. Hierarchal relationships among nouns can differ between them and keywords or features to improve the accuracy.

Our sample on online scientific sentiments domain. Consider the following example, “the author is a [lion-] in this field”, the previous review present negative polarity because it’s a name of animal but in real evaluation it’s a positive polarity. In the next review, “Bing is really [Einstein?]” evaluation sentiment analysis without world knowledge classifies above sentence as neutral, but it is an objective sentence because Einstein is the name of the famous scientist, so it refers a positive polarity also. It’s very hard for software to understand that automatically. The big problem in sentiments lexicon structure is huge lexicon to support the most words and their scores. But this research introduces a miniature lexicon. We can minimize the lexicon through two ways: the tree-structure and similarity and difference algorithms. This structure builds for each word based on two scores. The aggregate scores for each word equal 1. 𝑉(𝑤) = ∑(𝑊(𝑝) + 𝑊(𝑛)) = 1 V (w) refers to value of word, W (p) is to positive value and W (n) is a negative word, the selection between positive or negative polarity. Our proposed solution is based on an enactment bag-of-words (BOW) model [16] and with combine with Part-Of-Speech (POS) model. Enhancing BOW model cares of word level and evaluate word by word. But we use POS for caring a grammar and order of sequence of words. This lexicon constructs based on two phases: •Phase 1. Data Preparation Phase Less number of words in vocabulary lexicon to fast search based on similarity and differences algorithms. We neglect verbs tenses or word formula (singular or plural), that’s meaning we neglect English grammar and syntax because of our comparison and differentiation with the infinitive verbs, and singular words with most letters similarity. •Phase 2. Lexicon Development Phase Evaluation words is based on enhanced bag of words: we don’t depend on term frequency. It is based on assuming each word has two values and the total of them equal 1. Each term has 2 polarities (+/-). The construction of hierarchal database lexicon declare in figure.1:

Lexicon words

prefixes

Nouns keywords

Features

Nouns

topic

contribution

world knowledge

author

algorithm

unknown

In the following, we discuss the world knowledge lexicon solution flowchart: Figure.1: Proposed Lexicon Structure

A Database Solution for sentiments challenge

V. Input

EXPERIMENT

This experiment applies on two datasets: training set around 1000 reviews and human-verified sample set around 5000 reviews, by fitting to quantify the range with different sentiment analysis techniques can accurately evaluate polarity of text reviews. The experiment targets to improve accuracy [17]. The accuracy represents the rate at which the method predicts results correctly. The precision is the positive predictive rate, calculates how close the measured values are to each other. The F-measure results refers to the performance of the accuracy. Ideally, a polarity identification method achieves the maximum value of the F- measure, that is 1, meaning that its polarity classification is perfect.

For Each paper P do

Calculate Number N of Sentiments S

For each review R in P

For example: [The author of this paper look like Einstein]. Observation: The world knowledge is important challenge as in the previous review. [Einstein] in the name of scientist this refers to positive polarity but it is very hard to understand with computer algorithms. The results achieve nearly 70%. It improves accuracy and eases to understand. These results apply the two previous datasets, we discuss the accuracy analysis of world knowledge of improvement polarity levels.

For word w ∈ s do.

Remove Stop word list

Convert all words to upper case

Create a valuable Lexicon

74

If O (w) > 0 then.

Yes Yes

72.1

72

ACCURACY

No

70 68 66

65.6

64 If (W) has score?

62 accuracy

dataset1

dataset2

72.1

65.6

No Use Part of speech to recognize nouns only N o

If w is noun?

If w is keyword/fe ature?

Y es

N o Check world knowledge w

Detect review classificati

Detect sentiment Class (Positive, negative, V.P, V.N, neutral)

Get sentiment score for each w

Output

Figure.2: Flowchart of World Knowledge lexicon solution

Figure.3: Accuracy Percentage of World Knowledge related to Topic Domain

Our proposed technique introduces solutions to improve accuracy with the comparison between other two techniques, the world knowledge results achieve to 85% [2]. This results analyzed in figure.1. This results in scientific domain reviews improve on two datasets average results nearly 70%. But on the same results without the object domains.

A Database Solution for sentiments challenge

Accuracy Comparison 100 90 80

Accuracy (%)

70 60 50 40 30 20 10 0 0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Percentage of Datasets Model Figure.4: Accuracy Percentage any domain

The proposed solution cares of the grammar of each review sentences. It combines between bag-of-words model and Partof-Speech model. Our proposed lexicon relies on two parts: nouns and others. The others refers to any part of speech expect nouns such as verbs, adverbs, adjective, Etc. We can handle these parts of speech with using similarities and differences algorithms to achieve infinitive verbs. But the hardest issue in constructing lexicon is noun part. Noun has several types which have keywords, facts, other features. We discuss the close relationship and effect between the topic domain and world knowledge sentiment evaluation.

Presentation Layer

Application logical Layer Tokenizing Text analyzer

Sentiment classificati on

Summarizing

New proposed Lexicon

Sentiment evaluation

VI.

DESIGN AND IMPLEMENTATION

The proposed technique have used three layer architecture. The top most layer is the presentation layer (GUI), which manages all the interaction to end user. The middle layer is the application logic layer which includes all the functionalities such as text analyzer, sentiment classification, sentiment word evaluation techniques, lexicon which are used to manage knowledge resources. The bottom layer is the database layer and contains the database for paper, paper Metadata, and review relation and sentiment words and prefixes. The last layer refers to the solution of world knowledge challenge.

Database Layer

Paper

Paper_ Metadata

Review

Review_ relation

Sentiment words

Figure.5: Design of lexicon database

The implementation of our proposed technique using C# programming language working on Microsoft visual studio 2010 platform. the construction of proposed lexicon is based on SQL Server Management Studio 2008.

A Database Solution for sentiments challenge

V. CONCLUSION This paper explains the impact facts, knowledge information, and linguistics similarities. Our solution relies on a sentiment analysis word level. It introduces the proposed solution for several challenges in sentiment analysis to improve accuracy. It depends on the enhancement bag-of-words (BOW) model with combining POS model. Using POS supports our technique in caring of grammar and ordering of words. The technique applies on two datasets training and verified datasets for sentiment reviews. Experimental results show that our solutions nearly 70%, especially in scientific domain.

[1]

[2] [3]

[4]

[5]

[6]

[7]

[8]

[9] [10]

[11]

[12]

[13]

[14]

[15]

REFERENCES Bing, L., “Sentiment Analysis and Subjectivity”. In Nitin, I. & Fred, J. (eds). Handbook of Natural Language Processing. 2nd Ed, Machine Learning & pattern recognition series, Chapman& Hall/CRC, 2010. Doaa, M.E., "ASurvery of Sentiment Analysis Challenges", Journal of King Saud University: Engineering Science.April 2016. Doi:10.1016. Theresa,W. ,Janyce, W. ,& Paul, H., “Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis”, proceeding HLT’05 proceedings of the conference of Human Language Technology and Empirical Methods in Natural Language Processing, 2006. Hugo, L, Henry, L, and Ted, S, "A Model of Textual Affect Sensing using Real-World Knowledge". Proceedings of the 2003 International Conference on Intelligent User Interfaces, IUI 2003, January 12-15, 2003, Miami, FL, USA. ACM 2003, ISBN 1-58113-586-6, pp. 125-132. Miami, Florida, 2003. Subhabrata, M., Pushpak, B., "WikiSent : Weakly Supervised Sentiment Analysis Through Extractive Summarization With Wikipedia", Machine Learning and Knowledge Discovery in Databases, Volume 7523 of the series Lecture Notes in Computer Science pp 774-793. Hao, w., Dogan, C., Abe, K., Francois, B., and Shrikanth, N.,, "A System for Real-time Twitter Sentiment Analysis of 2012 U.S. Presidential Election Cycle", Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics, pages 115–120, Jeju, Republic of Korea, 8-14 July 2012. Doaa, M.E, Hoda, M.O., Osama, I., "Online Paper Review Analysis", Published in International Journal of Advanced Computer Science and Applications(IJACSA), Volume 6 Issue 9, 2015. Doaa, M.E., " Enhancement Bag-Of-Words Model For Solving The Challenges Of Sentiment Analysis", International Journal of Advanced Computer Science and Applications (IJACSA), January 2016. Archana, S., “Sentiment analysis of document based on annotation”, CORR Journal, Vol. abs/1111.1648, 2011. Walter, K., & Mihaela, V., “Sentiment analysis for hotel reviews”, proceedings of the computational linguistics-applications, Jacharanka Conference, 2011. Baccianella, S., Esuli, A. and Sebastiani, F., “SentiWordNet 3.0: An Enhanced Lexical Resource for Sentiment Analysis and Opinion Mining”, Proceedings of the Seventh conference on International Language Resources and Evaluation, 2010, pp. 2200-2204). Yu, Hong and Vasileios, Hatzivassiloglou, Towards answering opinion questions: Separating facts from opinions and identifying the polarity of opinion sentences, In EMNLP, 2003 12. M. Potthast and S. Becker, Opinion Summarization of 5. Web Comments, Proceedings of the 32nd European Conference on Information Retrieval, ECIR 2010, P. Turney. Thumbs up or thumbs down? semantic orientation applied to unsupervised classification of reviews, In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL’02), 2002 Neeraj, S., Liviu, P.,Raul, F.C., Abhishek, L., Chaitali, N., Adi-Cristina, M., Mallarswami, N., and Mirela, D., Database Fundementals ,First Edition, IBM, Canda, 2010 Pure Performance, Inc: Managing Hierarchical Data in SQL, 2012.

[16] Yin, Z., Rong, J., & Zhi-Hua, Z., “Understanding Bag-of-Words Model: A Statistical Framework”, International Journal of Machine Learning and Cybernetics, 2010. [17] Samih,Y., Erdogan,Y.,& Halife, K., “Tagging Accuracy Analysis on Partof-Speech Taggers”, Journal of Computer and Communications, 2014.

Doaa Mohey El-Din received her B.Sc. in Computer science from the Faculty of Computers and Information, Cairo University, Cairo, Egypt in 2010. She received her M.Sc in Information Systems at Faculty of Computers and Information, Cairo University, Cairo, Egypt in 2016. She is working toward his Ph. D degree. Her researches are focused on text mining, machine learning, and information retrieval. E-mail: [email protected]. She is a research assistant in information systems and machine learning field. She is a web developer and software engineer in Faculty of computers and Information, Cairo University. The previous work is a manager of training courses in information technology unit for visual & audio needs Unit, faculty of computers and Information, Cairo University, Cairo, Egypt. Mrs. Doaa, is a reviewer for several papers in (JASIST) Journal of the Association for Information Science and Technology, (JIPS) Journal of Information Processing Systems, and (JAL) Journal of Advances in Linguistics.