Flexible Searching and Browsing for Telecoms Learning ... - CiteSeerX

1 downloads 1159 Views 275KB Size Report
HTML, XHTML or other text format document. Three XSL templates were implemented to extract from the TopicMap all the knowledge relationships of a.
Flexible Searching and Browsing for Telecoms Learning Material Maria E. S. Mendes, Wayne Jarrett, Ognjen Prnjat, Lionel Sacks University College London, Dept. of Electronic and Electrical Eng., London WC1E 7JE, UK email: {mmendes | wjarrett | oprnjat | lsacks}@ee.ucl.ac.uk; Fax: +44 20 7388 9325 Abstract: In this paper, we describe the flexible searching and browsing tool for the telecoms learning material: the “Knowledge Navigator”. This tool was developed in the context of the European Commission sponsored project CANDLE (Collaborative and Network-based Distributed Learning Environment). Keywords: Searching and browsing, fuzzy logic, document clustering, metadata, XML, learning environment.

1. Introduction This paper discusses the design and implementation of the “Knowledge Navigator” tool, aimed at flexible searching and browsing of the networks and distributed systems (telematics) learning material. The tool was developed in the context of the European Commission sponsored project CANDLE (Collaborative and Network-based Distributed Learning Environment), under the Information Society Technologies (IST) program. The CANDLE project consortium consists of 11 technical universities across Europe and 1 corporate – British Telecom. The project aims at developing a methodology and a set of guidelines, supported by an appropriate pedagogical framework, for the development of network-based, multi-media telecoms and distributed systems courses enabling the re-use of teaching modules. The CANDLE approach is supported by developing a CANDLE system for authoring, maintaining, evolving, administering and brokering of multi-media, computer- and network-based courses, and the online access and navigation of the courseware. The Knowledge Navigator tool described in this paper is to be used by the Telematics students for searching and browsing of the CANDLE learning material. The learning material resides in the knowledge space composed of clusters of learning objects, which are related to each other with various strength relationships. The knowledge space is formed by an algorithm involving the fuzzy logic principles, which makes use of the CANDLE metadata used to describe learning objects. The approach developed here goes a step further from direct keyword search engines and flat knowledge space exploration by adding the concept of varying strength of relationship between different material and by grouping material in flexible clusters. An alternative solution would be to develop an ontology of the knowledge domain in question. An ontology consists of an explicit definition of a set of concepts and a set of relations between those concepts.

However, this approach has some limitations regarding the required maintenance effort and its subjectivity [Mend01a]. To overcome these limitations we have developed concepts for dynamic discovery of the knowledge space using fuzzy clustering. The approach presented here (in the context of the CANDLE project) was developed through a number of iterations, described in [Mend00] [Mend01a] [Mend01b] [Mend02]. In these, we have worked with a test collection of RFC documents (IETF standards). In [Mend03], the details of the fuzzy logic algorithm are discussed. The basic principles of the CANDLE project are described in [Sack02]. In section 2, we describe the CANDLE metadata schema which facilitates the creation of the knowledge space. Section 3 describes the structure of the knowledge space and its generation via fuzzy logic algorithms. Section 4 focuses on the design and implementation of the “Knowledge Navigator” tool, and section 5 discusses the deployment of the tool in the user (learner) trials. Section 6 concludes the paper and section 7 gives the references.

2. CANDLE metadata To facilitate the re-use of teaching material the CANDLE project has developed a methodology for modularization and tagging of the teaching material. Using this approach, the target granularity of the material (c-content) is established, and then the material is tagged with appropriate XML (eXtensible Markup Language) [XML] metadata. The CANDLE metadata represents an extension of the teaching and learning metadata developed through a number of initiatives, such as IMS [IMS], LOM [LOM] etc. The metadata captures a number of attributes which describe learning objects (or CANDLE c-content), such as general information (author, etc.), lifecycle information (version, etc.), technical information, rights information, etc. Specifically for CANDLE metadata, this also includes the details of the envisaged

pedagogical approach to be used, as well as metadata information concerned with classification of the material, where a number of well-defined keywords describe the content of the material. The reduced CANDLE metadata schema is depicted in Figure 1. The original schema was reduced to ease the manipulation of metadata.

discovered by a clustering algorithm based on fuzzy logic principles. Fuzzy clustering allows clusters to overlap by varying degrees, signifying that we recognize that the documents may be applicable to more than a single subject area depending on the context of the relationship, intended use, etc.

Cluster 1 D X

Cluster 2 Y

Cluster 3

Document D Belongs to: Cluster 1 = 0.5; Cluster 2= 0.5 Related to: Document X = 0.3; Document Y = 0.5

Figure 2 - Fuzzy knowledge space

The knowledge space that the tool gives access to is composed of a number of CANDLE c-atoms – units of learning material. The c-atoms are grouped together in clusters, with varying degree of membership of c-atoms in the clusters, and the varying strength of relationship between the documents in the same and overlapping clusters. This concept is illustrated in Figure 2.

There are several phases involved in the dynamic representation of the knowledge space. The fuzzy clustering algorithm takes as input the metadata keywords (from the reduced metadata schema, Figure 1) that describe the learning material (c-atoms), after being processed and encoded in a numerical form. The algorithm [Mend01b] that has been used is a modified version of the well-known Fuzzy C-Means clustering algorithm [Bezd81], which finds similarities between catoms and attributes them to one or more clusters with some degree of membership (in the range of 0..1). The knowledge space is then formed by extracting relationships between the c-atoms based on their common memberships is the various clusters. Thus, a learner can either directly search for particular material using the keywords, or explore the knowledge space freely by following the relationships (of varying “strength”) between the c-atoms or between the clusters of c-atoms.

The clusters are formed by grouping together c-atoms that share common topics of the knowledge domain. In Figure 2 we show three clusters, each of which classifies documents (c-atoms) within a certain subject area. Such relationships between c-atoms are

The clustering output is then represented according to the XML Topic Map standard (XTM) [Pepp01]. This standard has been recently developed to model and manage knowledge structures and information resources. The TopicMap model is very simple and

Figure 1 - Reduced metadata schema The classification part of the metadata – the keywords provide a basis for enabling free search, navigation and exploration of the teaching/learning material, performed by the learner. In the next section, we describe the process of forming the knowledge space, based on the CANDLE metadata and keywords.

3. Knowledge space creation

scalable. It consists of: topics (for describing knowledge “elements”, i.e. any set of subjects, ideas, concepts, etc.); associations (for specifying any kind of relationship between the topics); and occurrences (for enumerating which specific resources address a given topic). In the clustering TopicMap three types of knowledge “elements” or topics have been generated: the first type for metadata keywords, the second type for documents (c-atoms) and the third type for clusters. The fuzzy relationships obtained by the algorithm have been modelled as associations and the location of the actual XML metadata files have been included as occurrences of the c-atoms. One of the most direct uses of TopicMaps is for navigation and browsing information resources. In our case, this standard is used for navigating the dynamic fuzzy knowledge space of learning material. The browsing is facilitated through a simple user-friendly GUI which is described in the next section.

4. Knowledge navigator: design and implementation The “Knowledge Navigator” tool was designed for dynamic Web-based exploration of the fuzzy knowledge space. Such tool was required for interpreting the XML TopicMap and for displaying all the topics and associations within it in a meaningful way. The XSL standard (eXtensible Stylesheet Language) [XSL] has been chosen to transform the information contained in the TopicMap into an HTML frame set. XSL is not a markup language, but a sophisticated stylesheet formatting tool for XML documents. It consists of a series of rules (XSL Transformations) that form a “template”, which is used to format and/or modify the structure of an XML file, so it can be made recognisable to a Web browser or some other

application. XSL is extremely versatile, as it can sort, filter, add, delete and format XML data, and it can also transform an XML file into another different XML, HTML, XHTML or other text format document. Three XSL templates were implemented to extract from the TopicMap all the knowledge relationships of a given keyword, document or cluster, and to transform them into a frame set of HTML pages. The “Knowledge Navigator” was implemented as a Web server application based on Java Servlet technology. Documents, along with the XML metadata descriptions, were stored within a XML database and accessed through a Web browser. In response to user requests (HTTP requests), a Java Servlet runs the appropriate XSL template which then transforms the TopicMap for obtaining the desired relationship model. The HTML pages are generated dynamically and the document links are then populated with c-atoms by accessing the content database. All this is presented to the user through a simple GUI. In Figure 3a), a screen shot of the tool showing the relationships of a particular c-atom is presented. The current view refers to the “MSN Management – Overview” c-atom. By clicking on the link next to the document’s title (on the top frame), a new window opens, containing the document’s metadata – Figure 3b). Bellow the title frame, there is a list of Telematics topics that describe the document contents. The knowledge relationships of these topics can be seen if the links are followed by the user. On the left hand section of bottom frame (Figure 3.a) other documents that have a strong relationship with the selected document are listed. The right hand section of bottom frame reveals document groups of which the chosen document is a member. The weighted links assist the users in adjusting their position in the knowledge space should they need to search for other related material.

a) b) Figure 3 - The “Knowledge Navigator” tool: a) A view of the “MSN Management - Overview” c-atom’s knowledge relationships; b) CANDLE metadata for this c-atom.

5. User trials The use of the “Knowledge Navigator” was validated through a real-life trial where a group of 25 Telecommunications Masters of Science students were given an opportunity to revise for their final exam, over a period of one month, using the tool. The learning content presented to the students was the material from a lecture course taught on the University College London MSc. in Telecommunications programme entitled “Network and Services Management”. The lecture material was modularised and tagged with metadata using the CANDLE approach. Next, the material (c-atoms) was grouped in the fuzzy knowledge space, and the students were given an opportunity to revise the material by searching the knowledge space and browsing through the relationships between different topics. On average, the students interviewed spent 6% of their revision time using the Knowledge Navigator. Following is the summary of the results, where the scale is from 1 (not useful at all) to 5 (very useful). On average, those who have used the tool to find documents related to a specific topic and documents related to a specific document found the tool quite

useful (average scores 4 and 4.5, respectively). Some users found that the tool was only slightly less useful to find a specific document, whereas others did not find it useful at all for that purpose (average score 3.4). Regarding the relevance of the links, they found them more or less relevant to what they were looking for (average score 3.4). The feature that was most used to find documents was the keyword search facility (used very often with average score 4.8). Browsing by related topics and related documents were also used quite often (average scores 3.5 and 3.6, respectively). Browsing by document group was not used by one of them and only used sometimes by the others (average score 2). Regarding the usefulness of the document weights some users thought this feature was quite useful whereas others thought it was not very useful (average score 2.6). During their navigation, most users quite often came across other relevant material (average score 3.4). Only one user rarely came across other relevant documents. For more or less fifty percent of the cases, they were able to tell from the metadata whether a document was worth opening or not (average score 3).

Overall, the students particularly favoured the grouping and display of relevant documents, and the quick and easy search and location of documents. The students were highly in favour of including complementary material in the knowledge space, including that of other universities.

6. Conclusion In this paper, we have presented the design and implementation of the “Knowledge Navigator” tool, envisaged to be used for flexible search and browsing of the learning material. The tool was developed in the context of the IST project CANDLE. We described the CANDLE approach to decomposition and tagging of the learning material with the relevant metadata; the use of this metadata for the creation of the knowledge space through a fuzzy algorithm; and the approach to browsing the material, supported by the TopicMap standard and the user-friendly GUI. Finally, we have described the small-scale trial of the tool. The validation process indicates that the tool can be useful as an alternative to established direct keywords searches and “flat” explorations of knowledge space.

7. References [Bezd81] J. C. Bezdek, “Pattern Recognition with Fuzzy Objective Function Algorithms”, Plenum Press, New York, 1981. [IMS] IMS Learning Resource Meta-data Specification. http://www.imsglobal.org/metadata/index.cfm [LOM] Learning Object Metadata (LOM), IEEE standard for Learning Technology 1484.12.1 - 2002 ISO/IEC 11404. http://ltsc.ieee.org/wg12/index.html [Mend01b] M. E. S. Mendes, L. Sacks, “From Metadata to Fuzzy Knowledge Representation”. Proceedings of the London Communications Symposium, Sept. 2001, London. [Mend00] M. E. S. Mendes, L. Sacks. “Assessment of the Performance of Fuzzy Cluster Analysis in the Classification of RFC Documents”, Proceedings of the London Communications Symposium, Sept. 2000, London. [Mend01a] M. E. S. Mendes, L. Sacks, “Dynamic knowledge representation for e-Learning applications”, Proceedings of the 2001 BISC International Workshop on Fuzzy Logic and the Internet, FLINT 2001, Memorandum No. UCB/ERL M01/28, pp. 176-181, U. C. Berkeley, Aug. 2001.

[Mend02] M. E. S. Mendes, E. Martinez, L. Sacks, “Knowledge-based Content Navigation in e-Learning Applications”, Proceedings of the London Communications Symposium, 2002. [Mend03] M. E. S. Mendes, L. Sacks; "Evaluating fuzzy clustering for relevance-based information access", proceedings of the IEEE International Conference on Fuzzy Systems, Sept. 2003. [Pepp01] S. Pepper, G. Moore. “XML Topic Maps (XTM) 1.0”, TopicMaps.Org Authoring Group, Aug. 2001. http://www.topicmaps.org/xtm/index.html [Sack02] L. Sacks, A. Earle, O. Prnjat, W. Jarrett, M. Mendes, "Supporting Variable Pedagogical Models in Network Based Learning Environments"; Engineering Education 2002: Professional Engineering Scenarios, IEE, volume 1, 2002, Page(s): 22/1-22/6, London, UK, January 2002. [XML] Extensible Markup Language (XML) W3C Consortium, http://www.w3.org/XML/ [XSL] World Wide Web Consortium, “The eXtensible Stylesheet Language (XSL)”, http://www.w3.org/Style/XSL/

Suggest Documents