Towards realtime adaptation: Uncovering user models from experimental data

Abstract

Virtual Worlds and the non-player characters that inhabit them often lack knowledge about their users. Users are treated as sources of input or feedback. At best, systems respond to the user’s behavioural data captured in logfiles. But there is no deep understanding of the player. Without this deep knowledge it is not possible for the computer to intelligently adapt. Relevant knowledge about the user will differ according to the application domain. Currently studies capture data such as biographical details, health status and history, psychological profiles, preferences and attitudes via questionnaires. This data can not be used in real time to influence the behaviour of the system. We suggest that data collected in past studies could be used to create user profiles and rules that can be used in real time for tailored interactions. We present two examples in this paper, one relating to an educational virtual world for science inquiry and the other involving the use of an Intelligent Virtual Agent to reduce study stress.