Active User Modeling and Assistance with Dynamic Bayesian Networks

Intelligent user assistance systems face incomplete, uncertain and multiple modality sensory observations, user’s dynamically changing internal state, and many constraints in making decisions. We introduce a new probabilistic framework based on the dynamic Bayesian networks to dynamically and stochastically model the user’s internal state, profile, the sensory observations, and the contextual information. The framework provides a systematic mechanism that performs purposive and sufficing information integration to infer user’s state and provide correct assistance. This system works in a closed loop to fine-tune the assistance and to adapt the interface to the user. All these methods help to actively infer the user’s intention and engage appropriate assistance in a timely and efficient manner. A driver assistance system is designed as the application example.