Multi-agent framework for adaptive systems

Adaptive systems are expected to provide personalized services to a wide range of users with different interests, preference, skills and knowledge. Personalization functions in adaptive systems are usually assigned to user-modeling components (UMC), which must acquire user information into user models (through direct interaction with users or dynamically by analyzing user interaction patterns), resolve conflicts arising from model update (e.g. by overriding default stereotypical assumptions with concrete facts) and inferring additional knowledge about users. Apart from the UMC, an adaptive system may provide users with personal assistants (PA), which can autonomously perform user-delegated tasks. While the UMC and PA components may be lightly coupled within a monolithic adaptive system, architectures that support very loose integration of components are becoming attractive. Multi-agent systems provide an excellent framework for conceptualizing and implementing systems with loosely coupled components. We present in this paper a conceptual description of an adaptive system based on multi-agent framework in which a number of specialized agents interact to provide personalized services to users. An exemplar interaction protocol for the multi-agent system is provided using the FIPA-ACL communication acts. Finally, we discuss some of our ongoing work in this direction.