Abstract:
A major problem with incorporating a user model into an application has been the difficulty of acquiring the information for the user model. To make the user model effective, past approaches have relied heavily upon the explicit encoding of a large amount of information about potential system users. This paper discusses techniques for acquiring knowledge about the user implicitly (as the interaction with the user proceeds) in interactions between users and cooperative advisory systems. These techniques were obtained by analyzing transcripts of a large number of interactions between advice-seekers and a human expert, and have been encoded as a set of user model acquisition rules. Furthermore, the rules are domain independent, supporting the feasibility of building a general user modelling module.