Universität Würzburg

Phone: +49 931 888 6744

Fax: +49 931 888 6732

eMail: betz@informatik.uni-wuerzburg.de

April 17, 2000

The training of diagnostic problem solving capabilities is the aim of many case based training systems, e.g. for medical students. They follow the approach of problem based learning. This faces the students with a problem exceeding their capabilities and thus induces an information need. This paper examines different ways of accomplishing this need by generating hypermedia documents and enabling intelligent access to static documents.

Keywords: problem based learning, intelligent tutoring systems, hypermedia,

student modelling

D3Trainer is an intelligent tutoring system environment based upon expert-system toolkit D3. It is designed to teach diagnostic problem solving in different domains and is practical tested. In a biological domain students learn to classify plants. The system provides the student with appearance of blooms, leaves and stock and other features for examination. For the medical domain RheumaTutor and other systems has prooven the practical use. They confronts the student with virtual patients. Any action any time is possible for the learner. The student can select examinations, enter findings and diagnosis and select therapies. Based upon problem-solving capability of the underlying expert-system D3Trainer is always able to criticize the students actions. It can automatically generate explanations for these reviews. Therefore, the expert will not have to enrich each case with tests and reviews. Thus adding new cases takes only a few minutes (about 20 minutes for the RheumaTutor, including pictures). Unfortunately the development for the underlying knowledge base is somewhat more costly. So this approach is efficient only when applied to a large or constantly growing numbers of cases. Usage of the RheumaTutor perfectly fits this model being used in the education of medical students at the University of Munich. The medical history of a real patient is taken in one lesson. During the following session on the next day the students have to work with the RheumaTutor showing the same case as virtual patient. This is possible with a short authoring process only. At the moment a second version of the web-based client of the D3Trainer is under development. Web access eases updates of system and case base. This is essential for the wide-spread usage. It will center even more upon problem-based learning than the current version. Hence, it needs support for information generation or retrieval, presentation and navigation.

o Which examinations are reasonable?

o How should an examination be done?

o Which symptoms can be raised by an examination?

o Which list of diagnosis can be proven or confuted by a given examination

o Which therapy should be chosen at the moment and why?

o What are the effects of a therapy?

A lot of these questions can be asked with different aspects in mind. Given a diagnosis, one can ask for the symptoms supporting it. On the other hand, the question could also ask for a diagnosis supported by a given set of symptoms. Thus it is important for the query to know the origin of a question as well as its target. Information need is not only characterized by the question itself. Different people need different documents when asking the same question due to their different preferences and prerequisites. Several types of preferences can be thought of. For example student A chooses to inform himself by reading encyclopedias while student B prefers to read textbooks. Students’ previous knowledge is another important aspect of information need. Those with deeper knowledge need in-depth documents. They may be provided a more holistic view of the domain or compact documents summarizing the facts.

Not only dynamically generated information can be used in the current D3Trainer. The knowledge modeled within the system can also be enriched by preformulated hypertexts. These can be filtered with respect to the casecontext and the student model. An example is the survey of the medical history in rheumatology. When asking for the kind of pain, one can not ask for special movement patterns. Instead it is reasonable to ask for simple task, which normally cause the pain to occur. Teaching the formulation of these questions is not part of the knowledge-base. Instead, this information is associated to certain knowledge-base items. Some other subjects can not be linked to any one item in the knowledge base. Not any domain is closed up to a degree high enough to gather all important information in the knowledge base. To integrate the information on the border of the model, informal documents are to be interlinked with the knowledge base. Risks and side effects of therapies are good examples for this kind of information. They might interact with different domains — especially in a domain like medicine. Since this information may not fit properly to the knowledge-base, it might be impossible to connect it to some few items. A dynamic intelligent search is much better suited for finding these documents than browsing might be. But only by a guided retrieval process the student can get support. General purpose search engines based e.g. upon keywords are no proper way, since the student might get lost in all the less relevant data. A search including the students situation and the context of the problem will be able to provide the right documents to the student. This will be integrated in the D3Trainer be the use of a information broker architecture. Documents become enriched by meta-information (following e.g. the Dublin Core, [2]), connecting them loosely to the knowledge-base. With respect to different formats like HTML, XML or PDF, proxy objects are included. Hence the broker can handle different kinds of documents, because it will not have to look inside. Only the viewer will have to deal with the document itself. The proxy architecture allows the broker to reference documents somewhere on the world wide web. The search will be processed by an knowledge based approach. Due to the small number of documents (compared to internet search engines), a knowledge base for selection of the documents can be build. The problem-solving components provided by D3 can be used to classify the documents with respect to the users information need. Both browsing and searching can profit from information filtering. As discussed above, the students prerequisites influence the utility of a document. Both documents with to low and to high knowledge level are not helpful for the student. Thus these documents have to be filtered out in the first answer. With respect to the possibility of a wrong student model, the hidden documents should be shown on demand.

[1] De Vries, E. Educational multimedia for learning and problem solving,in Proc. European Conference on Artificial Intelligence in Education, Lisbon, Portugal, 1996

[3] Greer, J. and Philip, T., Guided Navigation Through Hyperspace, in Proc. Workshop "Intelligent Educational Systems on the World Wide Web", 8th World Conference of the AIED Society, Kobe, Japan, 1997.

[4] Puppe, F., Gappa, U., Poeck. K. and Bamberger S., Wissensbasierte Diagnose- und Informationssysteme, Springer, 1996

[5] Reinhardt B.,Kritikoptionen in fallbasierten Trainingssystemen in Proc. GMDS-Workshop in Heidelberg on Computer Based Training, Heidelberg, Germany, 1999

[6] Schewe S., Reinhardt B., Betz C. Experiences with a Knowledge Based Tutoring System for Student Education in Rheumatology in Proc. 5th German Conference XPS 99, Knowledge Based Systems - Balance and Perspectives, Würzburg, Germany, 1999

[7] Turtle, H. and Croft, B., Infrerence Networks for Document Retrieval, in Readings in Information Retrieval, Karen Sparck Jones and Peter Willet (eds.), 1990