Since the emergence of the World Wide Web (WWW) in 1991 (Berners-Lee,1991), the value of information has got a new dimension. Nowadays, millions of computers are connected via the internet, humans can collect information from nearly anywhere in the world, visit virtual galleries, go shopping at virtual market places, check their bank accounts or take a look at the actual situation at the south pole {http:bat.phys.unsw.edu.au/~aasto/}.

This enormous amount of information is also a chance for experience and learning. But effectively selecting information from the internet is still a hot research topic as the effectiveness of search machines increase with the precision of the query. The information contained in the internet is often useless for exploring or learning, as learners need guidance to build up a mental model of the area they are working on before being able to make sufficiently exact queries.

Promising approaches in research come from the area of adaptive hypermedia systems (Brusilovsky, 1996). Adaptive hypermedia systems combine hypermedia systems with intelligent tutoring systems. The aim of these systems is to personalize hypermedia systems to the individual users. Thus, each user has an individual view and individual navigational possibilities for working with the hypermedia system.

A contribution to adaptive hypermedia systems are adaptive hyperbooks (Henze, 2000) which personalize the access to information to the particular needs of users. They give users the ability to define their own learning goals, propose next reasonable learning steps to take, support project-based learning, give alternative views, and they can be extended by documents written by the learners. Adaptive hyperbooks have been developed in the KBS hyperbook project at the Institut für Rechnergestützte Wissensverarbeitung, whose English name (Knowledge Based Systems, KBS) gave the name for the project. Adaptive hyperbooks are information repositories for accessing distributed information. A main focus of hyperbooks is the extendibility of the system in respect to the World Wide Web. To create open adaptive hypermedia systems, the indexing approach chosen in KBS allows to treat each information unit equally independent of its origin. Thus, HTML pages from the World Wide Web can be integrated and adapted to a particular user's needs in the same way as documents stored in the hyperbook's library.

One of the main goals of student modeling in educational hypermedia is student guidance (Brusilovsky, 1996). Students have learning goals and previous knowledge which should be reflected by the hyperbook for adapting the content or the link structure of the hyper-document. For our KBS hyperbook system we follow a constructivist pedagogic approach, building on project based learning, group work, and discussions (Henze and Nejdl, 1997). Such a project-based learning environment leads to particular requirements for adaptation, in order to adapt the project resources presented in a set of hypermedia documents to the student's goals (for a specific project) and to the student's knowledge. It has to support the student learner by implementing the following adaptation functionality:
 

• Adaptive Information Resources: give the students appropriate information while performing their projects, by annotating necessary project resources depending on current student knowledge.
• Adaptive Navigational Structure: annotate the navigational structure in order to give the student additional information about appropriate material to explore or to learn next.
• Adaptive Trail Generation: provide guidance by generating a sequential trail through some part of the hyperbook, depending on the student's goals.
• Adaptive Project Selection: provide suitable projects depending on the student's goals and knowledge.
• Adaptive Goal Selection: propose suitable learning goals depending on the particular student's knowledge.

The user modeling component has to fulfill various tasks. On the one hand it has to enable the above stated adaptation functionality. On the other hand, it has to enable further adaptation functionality which depends on the openness of the KBS hyperbook approach (Henze and Nejdl, 2000): Information resources located anywhere in the WWW should be included in the curriculum of the student's work with the hyperbook, explanations and examples can origin from the hyperbook's libraries or from any other location in the WWW.

The connection between the KBS hyperbook system and the user modeling component is based on indexing any kind of information resources. The index concepts are called knowledge items (KIs). Knowledge items are similar to the domain model concepts used in (Brusilovsky and Schwarz, 1997) or the knowledge units in (Desmarais and Maluf, 1996).

 
3.1 Modeling the User's Knowledge

 
The knowledge of a user is modeled as a knowledge vector. Each component of the vector is a conditional probability, describing the system's estimation that a user U has knowledge on topic KI - on base ofall observations E the system has about U.
 

Definition 1:Knowledge Vector (KV(U))

where KI₁,.. , KI_n are the knowledge items of the application domain and E denotes the evidence the system monitors about U's work with the hyperbook. Observations about the student's work with the hyperbook are stored for each KI. Thus, the KIs are, on the one hand, concepts describing the application domain of a book, on the other hand , they are random variables with four discrete values coding the knowledge grades expert, advanced, beginner and newcomer. The evidence we obtain about the student's work with the hyperbook changes with the time. Normally, the student's knowledge increases while working with the hyperbook, although lack of knowledge is equally taken as evidence. Since every kind of observation about a student is collected as evidence, the knowledge vector gives - at each time - a snapshot of the student's current knowledge.
 

3.2 Indexing Information: HTML pages, Examples, Projects
 

Each information resource is indexed by some set of knowledge items describing the content of the resource. These resources can be general HTML pages, examples, projects, etc. The origin of an information resource is not relevant for indexing, only the content defines the index.
 

Definition 2:(Content Map)

Let S  be the (none-empty) set of all KIs, and let H be a set of HTML pages. Then

is the content map, which gives for each information resource in H the index of this resource, e.g. the set of KIs describing its content. P(S) denotes the power set of S. To identify the index of an information resource we can scan the text for keywords or phrases. Actually, the indexing is done by the author of an information resource by hand.

We have described the use of knowledge items for indexing all kinds of information, belonging to the hyperbook, or located anywhere in the WWW. The use of an indexing concept in student and user modeling in this way is new. Most approaches model dependencies like prerequisites or outcomes directly with the information resources themselves (Brusilovsky, 1999).We separate knowledge and information, as we model learning dependencies solely on the set of KIs of a hyperbook. The connection between the student modeling component and the hyperbook system is the content map (definition 2), which maps each information resource to a set of s.
 

Figure 1: Hyperbook Document "Methoden" (Methods) with links to examples, Sun Units and to two lectures where this document has been used

This separation is advantageous in many aspects. As the KBS hyperbook system allows different authors to write parts of a book, they become independent from the work of others: They can write (and index) their information entries without caring about the other content of the hyperbook. The KBS hyperbook system is an open hypermedia system, allowing to include information resources located anywhere in the WWW. As all information resources are equal in the sense that they only need to be indexed for being integrated in a particular hyperbook, this openness is enabled by the indexing concept, too. In addition, all kinds of information resources from arbitrary origins are fully integrated and adapted to the student's needs: We can propose programming examples in the WWW, generate reading sequences which contain material of the hyperbook library and the WWW, calculate the educational state of HTML pages in the WWW according to the student's actual knowledge state, etc. In fig.1 we can see an information page about Methods in Java. For this page links to examples, to the Java Sun Tutorial and to corresponding lectures have been generated in real-time and have been annotated with additional reading suggestions: A red ball indicates non-recommended information, a green ball indicates recommended information, and a white ball marks already known information.

In addition, the use of a separate knowledge model makes the hyperbook system robust against changes. If we add additional information pages or change contents, we only have to (re-)index these pages accordingly. No further work has to be spent on updating other material, as it would be necessary if knowledge, and thus reading or learning dependencies, would have been coded in the material itself.

We have proposed an approach for building open adaptive hypermedia systems on the Web. These open systems contribute to universal access as they allow to adapt information located anywhere in the internet to a particular user?s needs, goals and knowledge. Further work will concentrate on simplifying the discussed indexing approach by applying information retrieval techniques for (semi-) automatic indexing and retrieval of documents in the Web.

Berners-Lee, T. (1991): World Wide Web: An illustrated seminar. Held as an Online Seminar in 1991. http://www.w3.org/pub/WWW/Talks/General.html.

Brusilovsky, P. (1996): Methods and techniques of adaptive hypermedia. User Modeling and User Adapted Interaction 6, 2-3 (1996), 87-129.

Brusilovsky, P. and Schwarz, E. (1997): User as student: Towards an adaptive interface for advanced web-based applications. In Proceedings of the Sixth International Conference on User Modeling, UM97 (Sardinia, Italy, 1997).

Brusilovsky, P. (1999):Adaptive and Intelligent Technologies for Web-based Education. In Special Issue on Intelligent Systems and Tele-Teaching (Künstliche Intelligenz), Vol. 4, pp. 19-25.

Desmarais, M.C. and Maluf, A. (1996) : User-expertise modeling with empirically derived probabilistic implication networks. User Modeling and User Adapted Interaction 5 (1996), 283-315.

Henze, N. (2000) Adaptive Hyperbooks: Adaptation for Project-Based Learning Resources. PhD thesis, University of Hannover, Germany, 2000 . http://edok01.tib.uni-hannover.de/edoks/e002/313646791.pdf.

Henze, N. and Nejdl, W. (1997) : A web-based learning environment: Applying constructivist teaching concepts in virtual learning environments. In IFIP 3.3 and 3.6 Joint Working Conference: The Virtual Campus: Trends for Higher Education and Training (Madrid, Nov. 1997).

Henze, N. and Nejdl, W. (2000) : Extendible adaptive hypermedia courseware: Integrating different courses and web material. In Proceedings of the International Conference on Adaptive Hypermedia and Adaptive Web-Based Systems (AH 2000), Trento, Italy.