Fixed Probabilistic Evidence for Bayesian User Modeling

Abstract

Bayesian networks are an effective tool in diverse applicationsthat require reasoning and representation with uncertain data andknowledge. In fact, static models and traditional model learning are insufficientto translate the human machine interfaces. Thus, we employBayesian networks as a comprehensive and robust formalism to manageunknown information in user modeling. Inference is used to update thevariables within the Bayesian model when new information about othervariables becomes available. Evidence serves as the starting point for inferencemethods in Bayesian networks. This paper focuses on two distincttypes of evidence: hard evidence and probabilistic evidence. Our specificemphasis is on updating the evidence represented by a Bayesian model.The paper demonstrates the utilization of fixed probabilistic evidence,also known as soft evidence, in an adaptive user interface.