A framework for concept-based digital course libraries.

This article presents a general framework for building concept-based digital course libraries. The framework is based on the idea of using a conceptual structure that represents a subject domain ontology for classification of the course library content. Two aspects, domain conceptualization, which supports findability and ontologies, which support reusability of learning resources are incorporated uniformly in the library architecture. The suggested framework implies a layered information structure of the library content consisting of three layers, each capturing a different aspect of the information space--conceptual, resource-related, and contextual. We present an environment, TM4L (Topic Maps for Learning), aimed at supporting the development and use of concept-based digital course collections. The new ISO standard, XML Topic Maps (XTM), that provides a paradigm for organizing, retrieving, and interchanging information on the Web, is used in the proposed framework to implement concept-based digital course libraries.

**********

E-learning involves the use of new information and communication technologies and media to deliver, support, and enhance online learning, education, and training. The challenge it faces is to allow different modes of communication between learners and instructors, various ways of the provision of learning content, and enhanced support for online instructional activities. E-learning support systems' developers inherit from the achievements and lessons learned in the areas of traditional and intelligent pre-internet educational systems. However, the Web poses new challenges and opens new horizons to e-learning technologies. Typical e-learning scenarios involve preparing online instructional materials by instructors (authors) and extensive use of online learning materials by students (learners) in completing their learning tasks. Thus the success of e-learning applications depends crucially on the amount of effort that is involved in both developing educational materials by authors and retrieving relevant learning resources by learners. Important derivable questions then are: how to organize and classify the learning content and how to share, reuse, and exchange existing instructional units. In this article we discuss these questions in the context of one strategic category of e-learning applications, digital course libraries.

Digital course libraries are applications that contain instructional materials to assist students' learning in a specific discipline (course) and support students' course-related work (e.g. projects, assignments, etc.) that is aimed at reinforcing their knowledge. They play a vital role in out-of-class learning, and especially in project-based and problem-based learning. A good digital course library should provide learners with powerful and intuitive search tools that allow them to efficiently access sufficient amounts of relevant resources so that they can spend less time for searching and more time for learning. It should provide instructors with powerful authoring tools that support efficient creation, modification, and update of educational materials.

There are two main groups of problems related to using existing courseware, specialized educational collections, and educational portals intended to provide informational support for learning, as course digital libraries. The first group is related to the findability of online information: the information is often organized in such a way that the provided links are not intuitive and it is hard to find a needed piece of information. The broadly exploited hierarchical organization typically does not support "horizontal" navigation, that is, navigation to conceptually related information, relevant examples or different viewpoints. Another essential problem related to the efficient search of resources relevant to a specific learning task is that students are often unaware of the complete context of the task and need help in getting oriented in the conceptual structure of the subject domain. The second group of problems is related to the reusability of existing repositories of learning resources by instructors, which in turn concerns the existence of both shared agreement on their content and standards-based representation.

In this article we discuss the issues of findability and reusability of information in the context of digital course libraries. We address these issues by incorporating in the course library architecture a meta-layer--semantic layer, based on the 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.

Conceptualization can support topical finding of resources and learner's understanding of the specific subject domain (by enabling exploration of related domain concepts). Subject domain conceptualization has been long used for knowledge representation in intelligent tutoring systems (ITS). More recently, concept structures play a central role in adaptive hypermedia (AH) applications--for content fragmentation and structuring (Eklund, Brusilovsky, & Schwarz, 1997; Brusilovsky, Eklund, & Swartz, 1998; De Bra & Calvi, 1998; Murray, Gartner-Piemonte, & Kelleher, 2001), and in concept-based course support or learning support systems--for domain knowledge representation and information structuring (Vassileva & Deters, 1998; Greer et al., 1998; Aroyo, Dicheva, & Velev, 2001; Brusilovsky & Vassileva, 2003). These applications use the conceptual structure of a specific subject domain to organize digital material but they mainly support interconnected material and typically little or no references to external related documents. In the case of AH, the content is shareable (the generated dynamic web pages can be viewed in any browser), but not reusable and exchangeable. In the case of teaching and learning support systems, instructional materials have normally system-specific internal representation and therefore are not shareable, reusable, and interoperable.

The benefits of educational use of ontologies, however, have been only recently recognized (Mitrovic & Devedzic, 2002; Dicheva & Aroyo, 2002; Devedzic, 2003). Ontology is a well founded and broadly agreed upon system of concepts in a particular subject domain together with the relationships between those concepts. Thus specialized subject ontologies can be used as a semantic backbone for e-learning repositories. By providing common vocabularies for domain knowledge representation, ontologies are essential for instructional content "interaction." Courseware units can only be shared, reused, and exchanged among different authors if the authors agree on the vocabulary used in their construction. In addition, ontologies facilitate machine readability of web content. Ontology-aware digital course libraries can be implemented as ontology-based collections of standard learning objects (of right granularity) with standard technological support for describing, relating, retrieving, and presenting their content. In our framework these two different aspects, domain conceptualization, which supports findability, and ontologies, which support standardization and reusability, are incorporated uniformly.

In this article we propose a general framework for building efficiently searchable, reusable, and interchangeable discipline-specific repositories of learning objects on the Web. We also advocate using the new ISO standard--Topic Maps (Biezunski, Bryan, & Newcomb, 2000) that provides a paradigm for organizing, retrieving, and interchanging information on the Web, to implement concept-based, ontology-aware digital repositories.

GENERAL FRAMEWORK OF CONCEPT-BASED DIGITAL LIBRARIES

A key feature of our approach to digital course libraries is to use a network of concepts as both a medium of domain knowledge representation and a navigable structure. From one side, this would allow authors to create views of a specific domain in terms of domain concepts that suggest the semantics of the resources relevant to that domain--the view of interrelated resources is more than the sum of their parts. From another side, information retrieval would benefit from conceptual support based on domain ontologies since their predefined set of concepts, relationships, and inference rules constrain the possible interpretations. In order to be reusable and interoperable, learning objects must comply with technological standardization, in addition to knowledge standardization (consensus on the meaning of educational content). This implies that they should be represented using standard formalisms, including educational standards, such as LOM (IEEE Standard 1484.12.1-2002) and (XML-based) languages.

Thus the proposed general framework of concept-based digital course libraries is based on building a conceptual structure that represents a subject domain ontology and using it for structuring and classification of the library content. It is a continuation of our previous works on information handling support in concept-oriented web-based learning environments (Aroyo & Dicheva, 2001; Dichev & Dicheva, 2001). The classification involves linking learning objects (content) to the relevant ontology terms (concepts), that is, using the ontological structure to index the library content. This will allow applications and users to understand the relationships between the library resources and thus will insure efficient topical access to them. The use of subject ontologies that provide shared agreement on the subjects meaning will also allow for ontology-based merging of digital repositories.

Figure 1 represents the proposed general framework of concept-based digital course libraries. The main components are the information repository, the information-authoring module, and the information retrieval module. Detailed descriptions of those modules and their interconnections are presented in the next sections.

[FIGURE 1 OMITTED]

An architecture built within this framework uses the advantages of concept-based and standards-based content organization, which will benefit both learners and authors. For learners it will support efficient contextual retrieval of information relevant to their needs and for authors--the reusability, shareability, and interoperability of created instructional units.

Digital Course Library Repository

We propose a layered information structure of the library repository consisting of three layers (models), each of which captures a different aspect of the information space--conceptual, resource-related, and contextual (see Figure 2):

* Semantic layer, modeling conceptually the knowledge domain. This layer plays the role of a global schema, providing a declarative description of the subject domain in terms of key concepts and relationships among them.

* Resource layer, including a collection of diverse information resources associated with the specific knowledge domain.

* Context layer, representing different views (contexts) on the library resources depending on a particular application, goal, type of users, and so forth, by dynamically associating components from the other two layers. For example, a context can represent a collection of learning objects of a specific type (definitions, examples, demos, references, etc.) or a taxonomic view on the content in the form of a hierarchy of topics and subtopics, and so forth.

From an operational point of view, we define two operations on learning content (information structures):

* Aggregation, which enables users to build compound information structures.

* Fragmentation, which enables users to extract substructures from existing learning content.

These operations allow for a modular approach to learning content creation. The fragmentation operation is intended to generate a consistent subset of the original semantic structure along with the associated resources. The aggregation operation makes possible the creation of new learning content (a new compound structure) by combining separately developed learning components. It will allow instructors to combine content from different sources and to integrate previous results into new e-learning developments. The idea is to enable authors to create evolvable courseware with reusable components.

The suggested framework implies two steps of knowledge externalisation: a clear separation of the conceptual structure of the domain from the web resources, and an explicit (separate) representation of the relationships between information resources and concepts.

The semantic layer. One of the advantages of the proposed layered approach is that it enables us to discriminate a structure containing meaningful information about the resource layer. A valuable property of the semantic layer is that it exists separately from the information pool consisting of all learning resources. Therefore, apart from acting as a subject knowledge directory, the semantic layer can be used in its own right as a resource containing knowledge in the specific domain.

The introduction of a separate semantic layer representing the domain ontological conceptualisation entails two major benefits to learners. The first one is that it can support efficient context-based retrieval of information items relevant to learners' current goals by allowing natural and intuitive concept-based content browsing. The second one is that it allows for exploration of the ontological structure of the subject domain independently of the information resources, which can help learners to improve their overall understanding of the subject domain. According to Piaget (1974), building cognitive structures, called also "mental maps" schemes or "networked concepts," is an essential part of the process of understanding, responding, and learning the surrounding environment. The process of learning supported by concept-based courseware can analogously be seen as a continuous process of building and/or altering the learner's initial "mental map" through exploring a subject domain conceptual map and associated information resources (Aroyo, Dicheva & Velev, 2001).

[FIGURE 2 OMITTED]

The resource layer. The resource layer represents information resources associated with the specific knowledge domain. We impose some constraints to narrow our considerations to digital course libraries, that is, libraries containing online teaching and learning material. The information resources in this case are learning objects. We distinguish between internal and external learning objects depending on whether the object is stored locally in the library or is a web object. Internal resources are normally short pieces of information about a subject, such as definitions, characterizations, or short descriptions. Though external resources can be any addressable objects referenced by their URI, typically these will be standards-based (LOM, Dublin Core) learning objects. With good available collections of standardized learning objects, most of the resources will be specified as external, that is, by their URI, eliminating the need and efforts to create them. From this perspective, learning objects are another important source of reusability in our framework.

The proposed information structure is especially appropriate for supporting extensive exploratory work that requires guided research. For example, learning objects can be related to course work and connected with homework, project, and lab assignments. Guided research in this context assumes a sufficient set of resources coupled with an evocative conceptual layer providing meaning to the resources in terms of the interrelated concepts being learnt, that is, linking the resources to the right place in the knowledge map. For example, exploration of the resources related to indirect addressing and dynamic storage management can be guided by concepts such as "pointer," "dynamic variable," "dynamic memory," "dynamic memory allocation," "heap management," "garbage collection," and so forth, linked to each other by meaningful relations, for example, "dynamic variable" accessed by "pointer," "pointer" used for "dynamic memory allocation," "heap" kind of "dynamic memory," "dynamic memory" reclaimed by "garbage collection," and so forth. It also assumes access to richer repositories for the more interested learners.

The context layer. The separation between the semantic layer and the information repository makes it possible to define different semantic structures over the same collection of learning resources or different collections of learning resources connected to the same semantic structure. A context captures a particular view on the learning resources by preserving the relevant semantic relations among them (defined in the semantic layer) and filtering out the irrelevant. Thus the context layer can be seen as a collection of contexts, each one applicable in certain circumstances. By maintaining a collection of appropriate contexts, it is possible to categorize thematically the learning resources, reflecting multiple points of view. In effect, the context layer defines a set of navigable networks of learning objects that reflect different perspectives of their authors and define semantically customized views corresponding to different communities of learners.

The context layer opens the possibility of explicit representation of multiple contexts related to different situations and/or users' goals. Thus contexts enable search and navigation from an adequate perspective. This unlocks a new level for exploration of the digital content, which in turn facilitates learners in finding relevant objects. For example, one learner may prefer navigation through a specific sequence of course units (topics); another may prefer to explore learning content from a semantic (conceptual) point of view, and so forth. By supporting contexts, a user can access the same resources based on navigational strategies in conceptual spaces appropriate to his or her current needs.

The Layered Approach from an OOP Perspective

Interesting analogies can be made between the proposed layered approach for information modeling and the Object-Oriented Programming (OOP) model. The first observed analogies motivated us to try to extend them to more significant ones in an attempt to capture some advantages of OOP. The following list includes some of the analogies:

* The semantic layer in our model can be compared to classes in OOP.

* The context layer--semantic layer relationship can be compared to a subclass--superclass relationship in OOP.

* The external resources can be compared to instance variables in OOP.

* The internal resources can be compared to class variables in OOP.

* Merging can be compared to inheritance in OOP.

According to the OOP paradigm, reusability results from the possibility of creating different instances of a given class, coupled with the power of inheritance. Following these analogies, an important source of reusability in our framework is the ability to use the same semantic layer with different pools of learning resources coupled with the ability to create compound structures based on aggregation. Each semantic model (layer) defines the ontological structure of a knowledge domain. Thus the semantic models are somewhat less dynamic, more permanent and lasting than the learning resources they describe. Therefore, they can be viewed as reusable components of the framework defined in terms of "semantic-resource structure."

The context layer is intended to capture multiple views (including interpretations and classifications) on a particular collection of learning resources. Each context is a result of preserving or filtering out a specified set of relations on the resource collection defined by the semantic layer. In addition, a context can use some additional, application specific, non-ontological relations to link resources (e.g. relations of instructional nature). Thus we can draw a parallel between the semantic layer--context layer relationship and the OOP class--subclass relationship. A subclass can modify the behavior of a superclass by adding, changing or overriding some of the methods defined in the superclass. Analogously, a context can add, change, or override relations in the semantic layer. Therefore, the library collection can be viewed from different perspectives all of them being derived from the semantic layer by filtering out some relations and possibly adding some new relations. Each perspective presents a different sub-collection of resources--'instance' of a different context.

External resources are presented in the framework by links to web documents (outside the library) that provide additional information about given concepts. Since these resources are not embedded in the repository they can be easily changed with each new instance (by changing the corresponding links). Using these "adjustable" links to the semantic layer we can designate different external resources, depending on the intended application. This is in analogy with OOP instance variables that are associated with each individual instance of a class.

Internal resources provide a means for expressing short pieces of information about subjects. In contrast to the external resources, they are embedded in the repository and even information without a URI can be represented this way. They can be viewed as sharable pieces of information within the entire collection of instances of a particular structure. This is analogously to OOP class variables that can be shared amongst all instances of a class.

Continuing with our analogy, inheritance in OOP provides a form of code reuse by allowing one to take an implemented class and build a new class based on that implementation. Correspondingly, in our framework aggregation is intended to allow combining (i.e. reusing) separately developed learning units.

DL Information Authoring

We consider authoring digital course libraries a twofold process concerning both content development and content delivery. In the content development phase, the author has to create the digital content, that is, to define the resource and semantic models. In practical terms this involves defining the domain ontological structure and creating and linking digital library resources to that structure. Concerning content delivery, the author can specify choices related to configuring the final presentation of the digital content that reflect distinct adaptation to specific scenarios and individual users. This is related to the definition of the context model. Thus the digital library authoring tool supports (see Figure 1):

* semantic layer (ontology) authoring;

* resource authoring; and

* context authoring.

Considering semantic layer authoring, we expect that in creating the semantic layer authors will often use already existing subject ontologies. Further, we as other authors (e.g., Devedzic, 2003) anticipate that in the near future more and more subject ontologies will become publicly available (e.g., on educational servers) to be shared.

DL Information Retrieval

The second primary aspect of a concept-based digital library is the support for efficient retrieval of learning resources. According to the proposed framework, the information resources and concepts are separated in two distinct layers. This allows for support of two kinds of strategies for information seeking--opportunistic browsing strategies and analytical (i.e., key-word-based) strategies. Consequently, the digital library search tool supports the following (see Figure 1):

* Browsing: the user can navigate in the semantic layer through a series of related concepts (possibly based on a selected view) and inspect their associated resources.

* Keyword search: the user can make keyword-based searches in the resource layer (possibly based on a selected view) and inspect the concepts associated with the qualified resources.

* External search: the user can require external keyword searches (on the Web). This will provide important information about the completeness of the digital library content.

The task of supporting efficient retrieval for learning resources implies that the repository is able to meet the knowledge needs of learners. This involves keeping the information repository updated and ensuring that it contains enough learning resources on the topics of interest to users. Since this requirement is very difficult to meet, we propose to "open" the digital library to the Web and use learners' requests for searching the Web (external queries) as an indication for insufficient or missing information. Such feedback would help an author keep the repository continuously updated. For instance, if there are a number of external queries related to "pointers," "heap management," and "garbage collection," this would be an indication to the instructor that the information provided in the digital library in relation to those concepts is not sufficient or not presented at an appropriate level of detail or abstraction.

CONCEPT-BASED DIGITAL COURSE LIBRARIES: IMPLEMENTATION ISSUES 

Based on the general framework of concept-based digital course libraries, we are implementing an environment, TM4L, which enables the creation and contextual search of ontology-aware digital course libraries. The main design requirements to the environment include:

* The environment should have a scalable and expandable architecture.

* The environment should use XML-based schema language to create shareable, platform independent repositories.

* The environment should promote interoperability. It should have the capability to import/export structures from/to other XML--based repositories.

* The environment should support repositories "open to the Web," in other words, it should not only allow links to related collections residing at dispersed web sites, but should also support external searches on the Web.

The requirements were used as guiding principles for the selection of the development technology. We have chosen to use the ISO 13250 XTM standard--XML Topic Maps (http://www.y12.doe.gov/sgml/sc34/document/0323.htm) to implement the environment.

Topic Maps

Topic maps have their roots in traditional finding aids such as back-of-book indexes, glossaries, and thesauri. Unsurprisingly, this paradigm easily accommodates the representation and finding aids associated with digital collections. Topic mapping is knowledge representation applied to information management from the perspective of humans while RDF (http://www.w3.org/TR/REC-rdf-syntax/) is knowledge representation applied to information management from the perspective of machines (Pepper, 2000; Garshol, 2002). Topic maps can be viewed as a standard, interchangeable hypertext navigation layer above diverse electronic information sources that supports topical finding of various kinds of resources, such as documents, images, data-base records, audio/video clips, and so forth (Lin & Qin, 2002). The advantage of using the topic maps technology for developing digital learning collections is twofold: from one side it provides convenient and intuitive presentation of interrelated concepts embedded in information resources, and from another, the learning material is in a standard format, which makes it interchangeable. In addition, the topic maps standard defines the way in which two or more topic maps can be combined or merged. Merging is an important aspect of sharing topic maps that enables distributed creation of e-learning repositories and promotes reusability. Topic maps (as Semantic Web in general) are designed to facilitate navigating, searching, querying, filtering, customizing, sharing, and merging web information. The currently available commercial TM authoring tools, however, are too general and good for experts in knowledge representation but inappropriate for end users. While topic maps offer a powerful and promising technology for intelligent organization and access of information in general, creating topic maps for e-learning is not a clear and simple task at present (Dichev, Dicheva, & Aroyo, 2003). Therefore, specialised education-oriented TM tools are needed that facilitate the creation, maintenance, search, and visualization of topic maps-based learning resources.

As a TM illustration, Figure 3 presents a screenshot of the "List" topic of the AI topic map developed by us. It is displayed by the Omnigator--an application developed by Ontopia (http://www.ontopia.net/solutions/products.html) that allows loading and browsing topic maps.

All the topics associated with "Lists" are given under "Related subjects." The occurrences of the topic "Lists" (internal and external) are shown on the right organized by type. With internal resources (Application, History, and Definition) the contents of the resource are shown inline. With external resources (Examples and Notes), only the links to the resources are shown. Because "Lists" is a topic type, the Omnigator displays a list of topics of this type: "List Processing" and "Sets," in other words, an index of Lists subjects.

Topic Maps-Based DL Implementation

Topic maps provide appropriate and convenient platform for implementing our framework. The main topic maps components are topics, associations, and occurrences (Biezunski, Bryan, & Newcomb, 2000). The topics represent the subjects, that is, the things, which are in the application domain. They can have zero or more topic types and names (a base name and possibly variants for use in specific processing contexts). An association represents a relationship between topics. Associations have types and define roles of the participating topics. Occurrences instantiate topics to one or more relevant information resources. The scope feature defines the extent of validity of an assertion (a topic characteristic assignment): the context in which a name or an occurrence is assigned to a given topic, or the context in which topics are related through associations. An important concept in TM is this of identity. Two topics are the same if both have the same name in the same scope or both refer to the same subject indicator. The topics and all their characteristics could be merged if this condition holds.

[FIGURE 3 OMITTED]

It is clear now that the semantic layer in our framework can be easily implemented as a collection of associated topics. An important aspect of the topic maps associations is that they can exist despite the absence of occurrences linked to them. Further on, the resource layer can be implemented straightforwardly by defining topic occurrences. Figure 4 shows an excerpt from the XTM implementation of our AI topic map, including the definition of an association of type superclass-subclass between the topics "Prolog" and "Recursive rules" and specifications of an internal and an external resource for the topic "Lists."

One useful and potentially very powerful application of TM scope is to capture different views on information resources. Thus, contexts in our framework can be modeled based on the topic maps' scope feature. In addition, topic maps can be merged to provide a composite view of multiple sets of resources. Our aggregation operation can be modeled based on merging. By merging two or more topic maps, one can merge two or more (possibly different) views on the world, and still be able to differentiate between these views--to say that this is the subject as viewed by A for purpose X, and this is the subject as viewed by B for purpose Y. Fragmentation is not part of the current Topic Maps standard and will be provided as an additional functionality supporting a consistent extraction of a sub-map from the original topic map.

A number of research questions related to using topic maps for representing educational resources must be explored when designing TM-based educational tools, such as, how to translate the specific TM vocabulary into a domain-specific user-oriented terminology, how to personalise views on the same learning material, what are the "standard" views (if any) that might be of interest in different educational contexts, and so forth.

TM4L: A Tool for Servicing TM-Based Digital Course Libraries

The TM4L (Topic Maps for Learning) environment (in development phase) is aimed at supporting authoring and delivery (presentation) of learning material in a digital course library. Consequently, it consists of two major components, Educational Topic Maps Editor (TM4L Editor) and Educational Topic Maps Viewer (TM4L Viewer). The TM4L Editor allows authors to create, organize, edit, and reuse topic maps-based educational content. TM4L Viewer allows users to explore the content of the library through navigation or search.

TM authoring. The TM4L Editor is designed as a tool, in which authors externalize their subject knowledge in cognitive schemes represented as topic maps. The basic steps in building a topic map assume that the user has a conceptual specification of the learning content in terms of its topic set, association set, occurrence set, and their interconnections. However, creating such a conceptual specification is a difficult task and unfortunately there are no cookbooks on building topic maps. Therefore, specialized tools for supporting end users have to be developed in different areas, incorporating the specifics of the area to help users in defining topic maps in that area. According to our knowledge there are no specialized TM editors for the area of education developed so far. Thus our focus is on creating a specialized editor that improves and strengthens users' ability to efficiently create teaching and learning content based on topic maps.

The TM4L Editor offers specialized support for creating digital learning repositories. It benefits of the topic maps' basic feature to support easy and effective merge of existing information resources while maintaining their meaningful structure. The learning content created by the editor is compliant with the XML Topic Maps standard (XTM) and can be viewed by any standard TM browsing tool. To achieve interoperability, the editor will be able to import and export TM-based learning objects in XML format. In addition, it will provide extensive support to authors in developing topic maps, including:

* support for editing and modifying existing TM-based learning objects;

* easy comparison and merge of independently built TM-based learning collections;

* easy access to and manipulation of TM constructs (i.e., specific topics, associations, and resources) using a custom (education domain) language; and

* support for finding and adding external information resources, for example, through accessing key educational repositories, such as standardized vocabularies, encyclopedias, ontologies, and so forth, and educational portals.

The TM4L Editor's functionality includes the following capabilities:

* maintaining concepts: add concepts, delete concepts, link concepts to other concepts;

* creating learning objects: add learning objects, delete learning objects, modify learning objects, merge learning objects, define learning object metadata, link objects to concepts;

* creating contexts: define different views by linking learning objects, for example, conceptually, hierarchically, through applying certain instructional principles, and so forth; and

* import/export topic maps, that is, "transport" topic maps from one application to another.

The main objective in the design of the TM4L Editor was to minimize the burden of the content authoring on the instructor during the entire life cycle of the TM-based learning content.

TM browsing. The task of the TM4L Viewer is to support efficient retrieval of learning resources in the digital course library. The topics modeling the semantic layer's concepts represent the subjects, which are expressed by the resources. The semantic and resource layers are connected by occurrence links pointing from topics to resources. From the topic layer the user can access the needed resources. The fact that the learner can see the topic to which the currently inspected resource is related enables him or her to switch back and forth between the resource domain and the conceptual domain. The topics, providing the user with a hint where she or he is when looking at a resource, behave like satellites of the Global Positioning System (GPS) (Rath, 2001). Moreover, learners can use different contexts for navigation.

The TM4L Viewer's functionality includes the following capabilities:

* Browsing the e-learning repository:

- supporting orientation: showing the position during browsing;

- supporting perspectives: presenting information from different view-points;

- providing adaptive navigation depending on the selected viewpoint (for example, navigation following conceptual links or predefined instructional sequence of learning objects, etc.); and

- allowing change of navigation criteria at any time, e.g. shift to a navigation path in another view.

* Visualizing the structure of learning objects within a selected view.

* Keyword-based search for learning objects within the e-learning repository.

* Providing an option for external search on the Web.

The TM4L Viewer provides the course digital library interface that learners use. Therefore, the focus is placed on the intuitiveness and ease of its use in relation to library content browsing and search.

CONCLUSIONS

Our work is motivated by problems related to the efficiency of both courseware authoring and retrieval of learning resources. It is based on the idea of integrating concept-based digital libraries with subject ontologies that matches closely with the topic maps paradigm. We discuss a framework defining a novel architecture of topic maps-based digital course libraries, which is aimed at supporting efficient retrieval as well as reuse, exchange, sharing, and interoperability of digital learning materials. A secondary motivation behind this work is to evaluate the advantages of using the emerging topic maps technology for management of digital learning content in general and concept-based learning support systems in particular. Concept-based learning support systems represent a field of research and development that is highly regarded today and this work addresses an issue of the underlying dimensions that a concept-based approach can deal with in a rational and pragmatic manner. The proposed concept-based architecture of digital course libraries provides ground for efficient context-based retrieval of library information relevant to learners' current goals as well as for deeper understanding of the ontological structure of the specific subject domain. The proposed topic maps-based implementation of the architecture provides a unifying framework for standards-based knowledge representation and management, which promotes reusability, shareability, and interoperability of the learning content.

Future research directions include study of how to represent and use instructional knowledge in topic maps and how to model the learner and use this knowledge for 'library adaptation', etc. Another possible direction of future work is related to the information support necessary for the proposed 'opening' of a digital course library to the Web. We envision providing learners with contextual support for external searches on the Web, when the available library resources are not satisfactory for their goals. Some of the questions, related to this problem, include: What kind of (educational) context can be used to constrain the Web search and improve its quality and precision? Should we leave the learners in control of their search strategy or use a meta-search-like automatic modification of their search queries? A subject for future work is also to evaluate further the proposed environment based on a broad and longer-term experience of its use and to compare the results with other similar environments.

    <association>
    <instanceOf>
    <topicRef xlink:href="#superclass-subclass"/>
    </instanceOf>
    <member>
    <roleSpec><topicRef xlink:href="#superclass"/></roleSpec>
    <topicRef xlink:href="#Prolog"/>
    </member>
    <member>
    <roleSpec><topicRef xlink:href="#subclass"/></roleSpec>
    <topicRef xlink:href="#recursiveRules"/>
    </member>
    </association>

    <topic id = "lists">
    <instanceOf>
    <topicRef xlink:href="#Prolog"/>
    </instanceOf>
    <baseName>
    <baseNameString> Lists </baseNameString>
    </baseName>
    <occurrence>
    <instanceOf><topicRef xlink:href="#syntax"/></instanceOf>
    <resourceData>
    Notatiion: [XIXs]
    </resourceData>
    </occurrence>
    <occurrence>
    <instanceOf><topicRef xlink:href ="#examples"/></instanceOf>
    <resourceRef xlink:href="http://gorams.wssu.edu/faculty/dichevc/
      Listex.doc"/>
    </occurrence>
    </topic>

Figure 4. An excerpt from the XTM implementation of the topic map "AI"

Acknowledgment 

This material is based upon work supported by the National Science Foundation under Grant No. DUE-0333069 "NSDL: Towards Reusable and Shareable Courseware: Topic Maps-Based Digital Libraries."

References

Aroyo, L., & Dicheva, D. (2001). AIMS: Learning and teaching support for WWW-based education. International Journal for Continuing Engineering Education and Life-long Learning, 11(1/2), 152-164.

Aroyo, L., Dicheva, D., & Velev, I. (2001). A concept-based approach to support learning in a web-based course environment. In J. Moore, C.L. Redfield, & W.L. Johnson (Eds.), Artificial Intelligence in Education, 68, 1-10. Amsterdam, The Netherlands: IOS Press, Frontiers of Al and Applications.

Biezunski, M., Bryan, M., & Newcomb, S., (2000). ISO/IEC 13250 Topic maps: Information technology. Retrieved April 5, 2003 from www.y12.doe.gov/sgml/sc34/document/0129.pdf

Brusilovsky, P., Eklund, J., & Schwarz, E. (1998). Web-based education for all: A tool for developing adaptive courseware. Computer Networks and ISDN Systems. In Proceedings of Seventh International World Wide Web Conference, 30(1-7), 291-300.

Brusilovsky, P., & Vassileva, J. (2003). Course sequencing techniques for large-scale web-based education. International Journal for Continuing Engineering Education and Lifelong Learning, 13(1/2), 75-94.

De Bra, P., & Calvi, L. (1998). AHA!: A generic adaptive hypermedia system. In Proceedings of Workshop on Adaptive Hypertext and Hypermedia, (pp. 20-24). Pittsburgh, PA, USA.

Devedzic, V. (2003). Next-generation web-based education. International Journal for Continuing Engineering Education and Life-long Learning, 11(1/2), 232-247.

Dichev, C., & Dicheva, D. (2001, October). Deriving context specific information on the Web. In Proceedings of the World Conference on the WWW and Internet, WebNet'2001, (pp. 296-301). Orlando, Florida.

Dichev, C., Dicheva, D., & Aroyo, L. (2003). Using topic maps for e-learning. In Proceedings of the International Conference on Computers and Advanced Technology in Education (CATE 2003), (pp. 26-31). Rhodes, Greece.

Dicheva, D., & Aroyo, L. (2002). Concept and ontologies in WBES. In Proceedings of the Workshop Concepts & Ontologies in Web-based Educational Systems, (pp. 3-4). Auckland, New Zealand.

Eklund, J., Brusilovsky, P., & Schwarz, E. (1997). Adaptive textbooks on the WWW. In H. Ashman, P. Thistewaite, R. Debreceny & A. Ellis (Eds.), Proceedings of the Third Australian Conference on the World Wide Web, (pp. 186-192). Queensland, Australia.

Garshol, L.M. (2002). Topic maps, RDF, DAML, OIL: A comparison. Retrieved April 5, 2003 from http://www.ontopia.net/topicmaps/materials/tmrdfoildaml.html

Greer, J., McCalla, G., Cooke, J., Collins, J., Kumar, V., Bishop, A., & Vassileva, J. (1998). The intelligent help desk: Supporting peer help in a university course. In Proceedings of the International Conference on Intelligent Tutoring Systems--ITS'98, (pp. 494-503). San Antonio, Texas.

IEEE Standard 1484.12.1 (2002). Learning Object Metadata (LOM). [Online]. Available: http://Itsc.ieee.org/wg12/

Lin, X., & Qin, J. (2002, October). Converging knowledge structures into an integrated repository. The Journal of Library and Information Services.

Mitrovic, A., & Devedzic, V. (2002). A model of multitutor ontology-based learning environments. Proceedings of the Workshop on Concepts & Ontologies in WBES, (pp. 15-22). Auckland, New Zealand.

Murray, T., Gartner-Piemonte, J., & Kelleher, P. (2001). Evaluating features for conceptual and narrative flow in an adaptive hyperbook. In J. Moore, C.L. Redfield, & W.L. Johnson (Eds.), Artificial Intelligence in Education, Frontiers in Al and Applications, 68, 21-32. Amsterdam, The Netherlands: IOS Press.

Pepper S. (2000, August). Ten theses on topic maps and RDF. Retrieved April 5, 2003 from http://www.ontopia.net/topicmaps/materials/rdf.html

Piaget, J. (1974). Understanding causality. New York: Norton.

Rath, H.H. (2001). Semantic resource exploitation with topic maps. In Proceedings of the GLDV-Spring Meeting, 3-15.

Vassileva, J., & Deters, R. (1998). Dynamic courseware generation on the WWW. British Journal of Educational Technology, 29(1), 5-14.

DARINA DICHEVA AND CHRISTO DICHEV

Winston-Salem State University, Winston-Salem, NC USA

[e-mail]

[e-mail]