Data-Driven Refinement of a Probabilistic Model of User Affect

We present further developments in our work on using data from real users to build a probabilistic model of user affect based on Dynamic Bayesian Networks (DBNs) and designed to detect multiple emotions. We present analysis and solutions for inaccuracies identified by a previous evaluation; refining the model’s appraisals of events to reflect more closely those of real users. Our findings lead us to challenge previously made assumptions and produce insights into directions for further improvement.

• Type: Conference
• Year: 2005
• Where: UM
• Authors: Cristina Conati, Heather Maclaren
• Keywords: Dynamic Bayesian Networks, Probabilistic Model, Real Users, UM 2005