User Modelling and Hypertext Adaptation in the Tutoring System Anatom-Tutor

I. Beaumont

Fraunhofer Inst., St. Ingbert, Germany.

e-mail address: beaumont@ibmt.fhg.de


This article describes ANATOM-TUTOR, an intelligent anatomy tutoring system for use at university level. After a general introduction to the system, we briefly look at the system's question mode, in which explanations and questions are generated in a simulated examination situation, taking into consideration the individual user's level of knowledge, then take a closer look at adaptivity in the system's hypertext component, which also involves taking into account the user's level of knowledge.

General

ANATOM-TUTOR is, as the name suggests, a system for teaching anatomy. At present we deal with several aspects of brain anatomy, including the visual system, the pupillary light reflex system and the accommodation reflex system, each looked at from both morphological and functional perspectives. An intelligent knowledge base, accessible via menus and mouse-sensitive diagrams, is used for that part of the material which forms the framework of our domain. Detailed information, knowledge of how the information is used, and other material not suitable for representation in the knowledge base formalism is treated in a hypertext module. Much of this information is available (in rule form) to a didactic module which is thus able to present explanations and illustrate the use of the material. The didactic module adapts its presentation to the user's level of knowledge by consulting a rule-based user modelling component, which uses weighted rules and stereotypes to infer a default closure of the observed data complet ANATOM-TUTOR has three operating modes. In the browsing mode, menus and mouse-sensitive diagrams are employed for accessing the domain knowledge base. The user model is neither consulted nor modified in this mode. The question mode, which makes extensive use of the user model, is a simulated examination situation offering individually tailored responses to student misconceptions. In the hypermode, which also makes use of the user model, a hypertext based information system is used for the structured presentation of domain knowledge, together with illustrations of its use, in a manner suited to the users' level of knowledge. This is built along the lines of existing "stretchtext" systems such as MetaDoc from Boyle and Encarnacion [Boyle 91]. In the browsing mode, the user learns in a self-directed, explorative manner. In the hypermode, he is guided through the material in a personalised way, the important points are emphasized, and the uses of the material illustrated. The question mode can be used for exami It has been pointed out by several authors that Intelligent Tutoring Systems (ITSs) and educational hypertext systems are currently seen as two distinct approaches to using computers in education (e.g. [Brusilovsky et al, 94]). It would seem that this view has often tacitly implied the incompatibility of the two approaches. ANATOM-TUTOR shows that this is not the case, and that indeed the two approaches can be combined to the advantage of both. It is a system in which a hypertext component has been combined with an ITS, both modules referring to a common user modelling component for information about the learner. Its hypertext system is unique in that it has been equipped to offer custom-tailored explanations and auxiliary questions on encountering user misconceptions, a capability which relies on the properties of the ITS. Looking at things from a pedagogical point of view, ANATOM-TUTOR is, as far as the author is aware, the first comparable case of self-directed and guided approaches to instruction being simultaneously covered in one system.

The Question Mode

The question mode (a simulated examination situation) is important in that it provides the main vehicle for receiving information about the user. In any adaptive system, adaptation has to be carried out with respect to some (implicit or explicit) model of the user. This model has to be built and maintained, and the more complex the model, the more work has to go into filling it with information. In pure hypertext systems this is a problem, since information flows mainly in one direction - to the user. Many systems feature an enrollment procedure in which inventory is taken of the user's knowledge of the concepts covered in the material, but this can be extremely time consuming if a detailed model is required, and does not take into account the knowledge that the user gains by using the system. Updating the model on the basis of information asked for by the user is of limited usefulness. In ANATOM-TUTOR, the modelling problem is tackled as follows. The user first goes through an enrollment procedure. The aim here is to gather certain key information about the user, to which stereotype knowledge can then be applied. This is statistically gathered information about which items of knowledge are generally dependent on which other items. (For example in a programming application, one might find that users of a certain level who know the nroff command would also know the ls, mail, sed, etc., commands with some specifiable degree of certainty.) Using the user's knowledge of the key items, a default picture of the rest of his knowledge of the domain can be deduced. The actual questions provide a means of obtaining explicit information about the user's knowledge. If the user answers a question wrongly, but the user model contains the correct answer, his knowledge of the premises and domain laws necessary for deducing the information can be checked. This means that wrong answers result in relevant ad

The Hypertext Component

We will assume that the reader is familiar with the terms hypertext and hypermedia and will not give any basic definitions or descriptions here. Instead, we describe how the user model is used for tailoring text and specifying links, and how user's interaction with the system affects the user model. The idea behind ANATOM-TUTOR's hypertext component is to provide the learner with the full functionality of traditional hypertext with the additional advantage that the text has been adapted to the learner's individual needs, and that the user is not confused by having too many active links (lost in hyperspace). (This means, of course, that a detailed user model is required.) Adaptation in hypertext systems can take place at two levels: at the link level, and at the text level. Adaptation at the link level involves imposing some sort of structure on the nodes (or rather, on the concepts in them) and activating or deactivating links from these depending on the user's knowledge. In ANATOM-TUTOR, a hypertext lesson consists basically of pre-defined sequence of nodes. Depending on the global classification of the user, the actual nodes to be presented and the anchors to be activated are determined and the resulting setting remains unchanged during the course of the lesson. Adaptation at the text level can invol

In an information system such as hypertext, the main feedback which the system gets is the request for more information, which is usually done by clicking menu buttons or active fields in the body of the text. Boyle and Encarnacion, [Boyle 91], recognise that the mouse click allows only a "narrow bandwidth of information" but make the most of it by enabling the user to remove superfluous information (i.e. information which the user does not understand or which is too basic) from the screen, thus providing another feedback dimension. (We believe that in a realistic situation a user will just gloss over what he cannot use, as he would do when reading a book, and not try to use paragraph deletion as a signaling device.) If the user asks for an explanation of an item of information which was assumed known, MetaDoc removes the item from the user model and gives the requested explanation (presumably then re-adding the item to the user model). Paris [Paris 89] avoids the problem, and assumes the existence of a per While there may be a question of the desirability and user acceptance of putting test questions in an on-line documentation system, ANATOM-TUTOR is a tutoring program and we can use test questions to increase the bandwidth of information available from the user without leaving the teaching paradigm. Adding questions relating to the material is obviously superior to trying to accurately update the model solely from the user's requests for more or less material, and is in keeping with Self's [Self 88] advice on bypassing the intractability problem of student modelling - "avoid guessing - get the student to tell you what you need to know". The questions were designed with the task analysis approach to curriculum design in mind and are for confirming that the instructional objectives in the lessons have been attained. Not only do we get instant reliable feedback on the user's understanding, immediate active recollection is a recognised method of embedding information in the memory. When questions are posed by t

Implementational Details

ANATOM-TUTOR's user interface and hypertext component are programmed using XVT. The domain knowledge base, user modelling component, and didactic component are programmed in Common Lisp, the latter two components also using the Lisp based production system Ops5. ANATOM-TUTOR currently runs on Sun Workstations, and we are looking at the possibility of implementing a PC version.

References

[Beaumont, 94] Beaumont, I.: 1994, "User Modelling in the Interactive Anatomy Tutoring System ANATOM-TUTOR" in the journal User Modeling And User-Adapted Interaction

[Boyle, 91] Boyle, C., and A. Encarnacion: 1991, 'A User Model Based Hypertext Documentation System'. In: Proceedings of the IJCAI Workshop W.4. (The system is also described in: C. Boyle and A. O. Encarnacion: 1993, 'Meta-Doc: An Adaptive Hypertext Reading System'. In: User Modeling and User-Adapted Interaction.)

[Brusilovsky et al, 94] Brusilovsky, P., L. Pesin, and M Zyryanov: 1994 'Towards an Adaptive Hypermedia Component for an Intelligent Learning Environment'. In: Lecture Notes in Computer Science, Springer Verlag.

[Paris, 89] Paris, C.: 1989, 'The Use of Explicit User Models in a Generation System for Tailoring Answers to the User's Level of Expertise'. In: A. Kobsa and W. Wahlster: 1989, User Models in Dialog Systems, Heidelberg: Springer Verlag.

[Rich 89] Rich, E.: 1989, 'Stereotypes and User Modeling'. In: A. Kobsa and W. Wahlster: 1989, User Models in Dialog Systems, Heidelberg: Springer Verlag.

[Self, 88] Self, J.: 1988, 'Bypassing the Intractable Problem of Student Modelling'. In: Proceedings of the 1st Int. Conf. on Intelligent Tutoring Systems, Montreal.