Learning user models for an intelligent telephone assistant

User modelling is becoming ubiquitous in new "information appliances" and services. Mobile phones routinely include features such as predictive text input and voice recognition, tailored to a specific user. It is likely that, as with other AI technologies, user modelling will gradually become part of mainstream computing and products. The paper outlines a novel approach to user modelling by means of prototypes, implemented as support logic programs which exhibit typical behaviour. The user model is expressed as a dynamic distribution of supports over this set of prototypes. The new approach has been tested on a model application, the n-player iterated prisoner\´s dilemma, and on a telephone assistant system. Prediction success rates of over 80% have been achieved using simple prototypes, although further investigation is required to confirm the validity of this approach in the telephone assistant