Interactive Ontology-Based User Modelling

Speakers: Vania Dimitrova and Michael Pye from University of Leeds, Lora Aroyo and Ronald Denaux from TU/e
Date and location: 08.07.2004, TU/e, Eindhoven.

Visited and reported by Yulia Bachvarova and Katharina Schwarz

Vania Dimitrova (U of Leeds):

Vania presented a system called STyLE-OLM. This system can hold a dialog with a user to externalize the user's knowledge about the domain at hand. This way, the user's conceptualization of the domain is aligned with the ontology that the system uses and the user model is built. Both the system and the user can use the following rhetoric: The advantage of this over preconceived questionnaires is that the system can make sure that the user shares the same concepts, by probing deeper into a domain and finding out what a user knows about the concepts.
A remaining problem is the following: when a user offers something that is not represented in the ontology, it is classified as "wrong" by the system, although it might be correct. A better way to deal with this would be to keep these statements and check whether contradictions occur, and save them until the next evaluation of the ontology.
A further disadvantage is that this process takes quite a long time. Therefore STyLE-OLM will be integrated with AIMS of Lora Aroyo for rapid user modeling.

Lora Aroyo (TU/e):

Lora's system AIMS stands for Adaptive Information Management System. It is a task-based information retrieval environment that uses a domain model, a resource model and a task model.

Joint project (U of Leeds and TU/e):

The project aims to contribute to the initiative of creating Educational Semantic Web.
They want to integrate the dialogue functionality of STyLE-OLM to support the building of a user model based on the domain ontology, but the process should be faster. As Yulia observed, this seems to be a contradiction, because externalizing a user's knowledge via dialogue is usually a lengthy process, even between two human beings. As Katharina understood it, it is just relatively faster, in comparison with the system STyLE-OLM.
STyLE-OLM was developed using conceptual graphs and rules executed in Prolog. With the new project everything will be translated and executed in OWL DL. They still don't know to what extent this will be possible.
In STyLE-OLM as well as in AIMS graphical communication with the user was mainly employed. This approach will be kept in the new project. The person responsible for its development will be Michael Pye. He presented the existing tools for visualizing ontologies and some first ideas about interface of the system the project attempts to develop.

Discussion

Yulia: During the discussion it was pointed out that visualizations of semantics spaces normally show only a small part of the whole semantic network and the user loses the whole picture. On the other hand, showing the whole picture always causes information overload. It was suggested that when a specific smaller semantic space is visualized and there is the need to show a relation to a distant concept (because some inference has to be illustrated, for example) this concept can be "moved" (this transition can be graphically illustrated) to appear "in focus" so its relationship with the existing representation is also seen.

Katharina: I missed an ultimate goal of the project, because what they presented showed how they intended to build the user model, but not what they were going to do with it. As Paul de Bra pointed out, the scenario they described, which was teaching all about Unix, seemed more like a help system than a learning environment. The system would respond directly to specific tasks that users presented and explain how to solve the tasks at the level of the user, be it beginner, intermediate or advanced. It was completely tailored to what the user wanted to know, but there was no direction such as a certain level the users had to reach at the end of a course.

Finally, we took several handouts from the talk for those who are interested in more detail.