Assessing Temporally Variable User Properties With Dynamic Bayesian Networks

Bayesian networks have been successfully applied to the assessment of user properties which remain unchanged during a session. However, many properties of a person vary over time, thus raising new questions of network modeling. In this paperwe characterize different types of dependencies that occur in networks that deal with the modeling of temporally variable user properties. We show how existing techniques of applying dynamic probabilistic networks can be adapted for the task of modeling the dependencies in dynamic Bayesiannetworks. We illustrate the proposed techniquesusing examplesof emergency calls to the fire departmentof the city of Saarbrücken.The fire departmentofficers are experienced in dealing with emergency calls from callers whose available working memory capacity is temporarily limited. We develop a model which reconstructs the officers' assessments of a caller's working memory capacity.