Alexander Nikov*, Wolfgang Pohl**
* Technical University of Sofia, PO Box 4 1, BG-1612 Sofia, Bulgaria, email:
firstname.lastname@example.org. **GMD-FIT, HCI Research Group, Schloss
Birlinghoven, D-53754 St. Augustin, Germany, email: email@example.com
User-adapted interaction has been a research goal for many years now. In research on natural-language dialog systems, user models were introduced as explicit representation of assumptions about the individual characteristics of users. Systems would consult these models to decide how to adapt their behaviour to each user. Traditional user models contain mentalistic assumptions (MA), i.e., assumptions about knowledge, goals, interests and other mental attitudes. They are typically represented explicitly in some symbolic or numeric format ( Pohl 1998). The main problem in this traditional approach is how to acquire relevant assumptions about the user. In most systems, heuristics were used to control user model acquisition, which often led to unreliable system adaptivity.
In recent years user-adaptive systems like interface agents and personal assistants ( Maes 1994) have been developed that use a different approach: They analyse observations of user behaviour applying machine learning techniques, and adapt to the user based on usage patterns detected. Thus, these systems form behaviour-oriented assumptions (BOA) about the user in a systematical and reliable way, but without representing them explicitly (see Davidson and Hirsh 1998 for a recent example). Hence, the user cannot inspect and control the assumptions the system holds. Moreover, mentalistic assumptions needed by many adaptive applications are not constructed.
Our idea is to show that both mentalistic user modelling and behaviour-based usage modelling can be beneficially used to develop user-adaptive systems. We propose the MBAUM (m+̱entalistic and b+̱eha+̱viour-oriented u+̱ser m+̱odelling) framework that combines both approaches and is based on ideas described in