Self-Adaptation of a Learnt Behaviour by Detecting and by Managing User's Implicit Contradictions

This paper tackles the issue of ambient systems adaptation to users' needs while the environment and users' preferences evolve continuously. We propose the adaptive multi-agent system Amadeus whose goal is to learn from users' actions and contexts how to perform actions on behalf of the users in similar contexts. However, considering the possible changes of users preferences, a previously learnt behaviour may become misfit. So, Amadeus must be able to observe if its actions on the system are contradicted by the users or not, without requiring any explicit feedback. The aim of this paper is to present the introspection capabilities of Amadeus in order to detect users contradictions and to self-adapt its behaviour at runtime. These mechanisms are then evaluated through a case study.