Topic Maps-Based Digital Libraries

Digital course libraries are educational Web applications that contain instructional materials to assist students' learning in a specific discipline. They play a vital role in out-of-class learning, especially in project-based and problem-based learning, as well as in lifelong learning. Digital course libraries are expected, on one side, to provide learners with powerful and intuitive search tools that allow them to efficiently access learning resources, and on another, to support instructors with powerful authoring tools for efficient creation and updating of instructional materials. The latter is closely related to the issue of reusability and shareability of learning content, which in turn is related to both the existence of shared agreement on the content and the standards-based representation of the materials.

We address the problems of findability, reusability, and shareability of learning materials in digital course libraries by suggesting the use of Semantic Web technologies in creating them. More specifically, we propose a framework for digital course libraries that incorporates a meta-layer - semantic layer, based on conceptualization of the course subject domain. The fundamental idea is to build those libraries as both concept-based and ontology-aware repositories of learning objects. Further on, we propose that the implementation of such libraries is based on the ISO XTM standard - XML Topic Maps. Topic Maps (TM) are an emerging Semantic Web technology, that can be used as a means to organize and retrieve information in e-learning repositories in a more efficient and meaningful way.

Topic Maps provide an external meta-structure (a knowledge navigation layer) in the form of a dynamic, semantically based hypertext. As a result, TM-based courseware can offer the following benefits:

Currently available commercial TM software is mainly aimed at supporting rapid development of TM-based applications (e.g. Ontopia Knowledge Suite, Mondeca Intelligent Topic Manager, etc.). There are some available TM authoring tools, but they are useful to experts in knowledge representation, not to end users. We are not aware of existing specialized education-oriented TM tools that can be used to facilitate the creation, maintenance, search, and visualization of Topic Maps-based learning resources.

TM4L is an e-learning environment, which enables the creation, maintenance, and use of ontology-aware learning repositories based on Topic Maps. It provides support in conceptual structure design and maintenance through its functionality for editing, browsing, and combining such structures, coupled with support for relating concepts, linking concepts to resources, merging ontologies, external searching for resources, defining perspectives, etc. The TM4L environment consists of a TM Editor and a TM Viewer.

The TM4L Editor is an ontology editor allowing the user to build ontology-driven learning repositories using Topic Maps. It provides ontology and metadata engineering capabilities coupled with basic document management facilities. The TM4L Editor benefits from the Topic Maps’ fundamental feature to support easy and effective merge of existing information resources while maintaining their meaningful structure. This allows for flexibility and expediency in re-using and extending existing repositories. The learning content created by the Editor is fully compliant with the XML Topic Maps (XTM) standard and thus interchangeable and interoperable with any standard XTM tools.

The TM4L Editor is Topic Maps-based, thus the main objects that it manipulates are topics (representing domain ontology concepts), relationships between them, resources, and contexts (represented by themes). It includes four different sections (views): Topic Map, Topics, Relationships, and Themes. The user interface uses the Tab metaphor; each tab is associated with a different view on the Topic Map: Topics, Relationships and Themes view. Screenshots from the TM4L Editor interface (the Topics and Relationship sections) is shown on Fig. 1.

Figure 1. Screenshots from the TM4L Editor interface: Topics and Relationship sections.

Topic Map. In the Topic Map section the author defines metadata (Dublin Core and LOM compliant) for the newly created Topic Map. This includes: TM Title, Creator, Subject / Main Topic (keywords), Description, Publisher, Contributor, Creation Date, Last Modification Date, Language, Location, Source, Relation, Coverage, IPR / Copyright. Additionally, a Topic Map Subject Indicator is specified. Some LOM tags are automatically included in the TM metadata with pre-specified values, e.g. LOM 4.1 Resource Format (“text/html”), LOM: 5.1 Interactivity Type (“expositive document”), LOM: 5.3 Interactivity Level (“high”), etc.

Topics. In the Topics section the author defines, edits, and deletes topics. Each topic definition includes the following information: subject indicator, names, types, and related resources. For each new topic an ID is automatically generated.

Topic categories. Our major concern in designing the Topic Maps Editor was related to the fact that in the TM standard every subject is a topic, which is a powerful idea but will not make much sense to the uninitiated authors. Three different kinds of topics are expected to be used in an educational Topic Map: ‘concept’ topics needed to build the ontological representation of the specific subject domain, ‘utility’ topics needed as meta-data fillers in the Topic Map, for example, to specify the different types of educational resources, and ‘system’ topics needed to represent association types, roles in associations, and other entities required by the TM model. In TM4L we combine the utility and system topics and support two distinct categories of topics: domain ontology topics and utility topics. The former are defined by the user and listed in the Topics section; the latter are automatically defined by the editor, e.g., when a specific authoring activity (such as defining a new relationship type) takes place and are not normally listed in the Topics section. We use the following utility topics categories: association types, association role types, occurrence types, name use types, and themes (for scoping). The category of a topic depends on where it was created by the user, for example, if it was created as a result of user input in the ‘Create Relationship Type’ dialog, it is an association type.

Topic names. TM4L allows multiple topic names: one primary and possibly some alternative names. Each name can have alternate names (TM name variants) to be used for special purposes. In this application we have constrained the number of alternate names to four, corresponding to four different purposes of usage of the name: sort, search, display, and draw.

Topic Types. In compliance with the XTM standard, multiple topic types are allowed. The user is given two ways to declare a topic type (or parent topic): either automatically by selecting an existing topic prior to the creation of the new topic, or manually by adding a parent in the ‘Parent Topic Panel’.

Resources. Resources can be internal and external. Internal resources are short pieces of information about a concept, such as definition, short description, some characterizations, etc., stored locally in the Topic Map. External resources can be any addressable learning objects on the Web referenced by their URI. For authors’ convenience, some resource types are pre-defined however the author is allowed to define their own types. We have predefined the LOM 5.2. Learning Resource Types: exercise, simulation, questionnaire, diagram, figure, graph, index, slide, table, narrative text, exam, experiment, problem statement, self assessment, and lecture. In addition, we have predefined types of learning resources relevant to characterizing subject domain concepts: definition, description, example, and graphical representation.

Relationships. Relationships in our model are represented by Topic Map associations. Each relationship has a type (e.g., ‘is-component-of’) and one or more members (concrete topics) along with the roles they play in the relationship. There is a pool of pre-defined relationship types (such as ‘class-subclass’) that the authors can use. In the Relationships section of the Editor the author can define relationship types and roles, create relationships by specifying their types, roles, and role players, and edit and delete relationships. When defining relationships the author selects all involved entities – relationship type, members, and roles, from presented lists, so that input errors are minimized. The scope (context) within which the assertion made by a relationship is valid can be defined in the Theme section. If none such is present, the scope is unconstrained and the relation is always valid.

Instead of adopting a single “perspective” on classes of concepts, our model includes three basic concept hierarchies. In this way we are able to create more expressive conceptual structures that include various classifications of certain concepts. For example, operators can be classified by arity (unary, binary, and so on) or by type (arithmetic operator, Boolean operator, String operator, and so on); Java threads can be classified as “part-of” the JavaVM or as “sub-class” of the user–level threads. By enabling different perspectives, we can model different classifications of topics at the same time

Contexts (Views). We conceive the notion of context as derived from two principles: the principle of grouping and the principle of locality. According to the first principle the context is a notion, capturing a fragment of related entities. Grouping can be based on different criteria and different assumptions. We perceive context as a generalization of “putting together”, clustering and categorization: it combines all types and patterns of grouping. That’s why the notion of context is so elusive, despite of the various models proposed and developed. According to the second principle the notion of context is an abstraction, capturing the localization principle in various aspects. For example the principle of locality applied to a given topic in terms of the topics directly or indirectly related to it, determines the set of topics that are somehow related to the selected topic, that is, determines the context of relevant topics. If we apply the localization strategy towards selected relation type, then we arrive to a new type of context (perspective). For example if we select the “whole-part” dimension we can see the topics from a particular hierarchical perspective. Thus the proposed approach to contexts captures also the notion of perspectives (or viewpoints).

TM4L allows authors to define contexts through the use of relations and scope (theme). The notion of theme makes it possible to express multiple viewpoints on a single set of learning resources and provide personalized views for different groups of learners. The theme mechanism of TM4L enables any information provided about a topic to be qualified by defining a context within which the information is valid. Theme may be used to define several different perspectives on the same set of information. For example, theme may be used to separate "beginner" resources from "intermediate" or "advanced" resources, thus enabling different sets of information to be presented to learners on different levels.

The TM4L Editor is implemented in Java and uses the TM4J Topic Map Engine, which is an open source providing a comprehensive API that allows creating and modifying Topic Map structures stored either in-memory or persistently in a database. The Editor has open modular architecture that allows an easy extension of its functionality.

We consider the exploration practice as the process of finding information that is relevant to the learner’s current tasks. There is a tendency towards browsing in terms of exploration, and the TM4L Viewer should therefore be enhanced to better support both browsing and the combination of search and browse activities. The exploration practice differs from information querying in that no specific question needs to be answered. Instead, the user/learner wants to know about relevant information at a more global level, e.g. to see what information is available in terms of their current information needs. Exploration also differs from general analysis in that the issue is not to oversee the entire collection in a holistic way but only inspect those parts relevant to the learner’s current task. The exploration of large information spaces is a difficult task, especially if the user is not familiar with the terminology used to describe information. Conceptual models of a domain in terms of thesauri or ontologies can remedy this problem to some extent. Exploration on the level of concepts and relationships can be used as a navigation and query formulation mechanism fostering semantic exploration and discovery. In order such an ontological framework to be useful, there is a need for interactive tools for exploring large information sets based on conceptual knowledge.

By the term “informativeness” we refer to the extent and type of information that the concept structure should include in order to satisfy the information needs of the envisaged learners. In this respect, we consider a concept structure as informative if it provides a representation of the properties of the involved concepts along with the relations between concepts that is at an appropriate level for the learners to understand.

The predefined relation types for our purposes were selected on the basis of pragmatic considerations. A key consideration was the information overload, related to both the amount of information provided and the ability to process it. In particular, information overload can result from a concept structure that is too complex (in terms of relations among the concepts), or from a non-carefully designed graphical user interface. The set of relation types included as predefined relations was selected partially on the basis of the expected background of our envisaged users. It is relatively small but can be extended with arbitrary number of associations based on the TM standard. Three of the hierarchical relations are the most commonly studied and used semantic relations (although slightly adapted to our purposes), “superclass-subclass”, “whole-part” and “class-instance”; the other relation types have been selected especially for capturing concept structures typical in the intended domain: “related-to” and “similar-to”.

These observations suggested in turn the following guiding principles with respect to the TM4L Viewer design:

To enable multi-purpose exploration TM4L supports multiple views. Interfaces that provide multiple views offer users different perspectives on a selected entity. TM4L visualization strategy is to provide view on demand.

TM4L has been developed as a general information course-task support tool. Therefore, it has a general user interface, not dependent on a specific knowledge area. The goal is to provide an intuitive graphical interface for Topic Map-based learning content navigation. Three views are currently supported by the Viewer: Graph View, Tree View, and Text View. These views are intended to ease navigation at “hot spots”. The graph view includes a semantically expressive, browsable graph (based on HyperGraph) (see Figure 2).

The interface allows browsing all the topics and relationships defined in the Topic Map as well as filtering some views with respect to selected topic types or relationships. The visual display is not intended to convey the full richness of a TM-based repository, but to show which topics are present and how are they related. Aiming at reducing the information overload, we have chosen at each navigation step to display only the topics most immediately related to the currently selected object. In addition, we have chosen not to show the resources associated with the displayed topics in the Graph view, since the visualization becomes too crowded and unclear. Thus the Graph view represents only ‘ontology’ objects - topics, relationships, roles (the latter can be also hidden) but not resources.

The TM4L Viewer supports this by offering six corresponding indexes. These indexes provide the starting point for browsing the Topic Map. When the user selects in an index a particular object (topic, relationship, etc.), it becomes the “focus” object and will be displayed in the “Tree View” and “Graph/Text View” panels of the Viewer’s window. The view in each panel can be changed to any of the other two. The user can continue browsing the learning content by selecting an object related to the currently displayed one. When navigating, the user can choose in which panel the information about the selected topic is to be displayed. This allows browsing different objects related to the current one without loosing the focus. By exploring the graph in a particular direction the user can obtain a better understanding of its content and thus decide what portion of the repository is relevant to their needs.

The user interface displays only small portions of the Topic Map objects at any time. The TM4L Viewer provides an animated and zoom-able view with context sensitive features like click-able topics or selective detail views.

The current version of the TM4L Viewer is a result of prototyping of different visualization ideas that offered us a rich design alternatives. Its implementation is based on TMNav, which is part of the TM4J open source project.

This material is based upon work supported by the National Science Foundation under Grant No.0333069. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

Towards Reusable and Shareable Courseware: Topic Maps-based Digital Libraries