An Affective Virtual Agent for Natural Human-Agent Interaction
1. Matthijs Pontier, Marco Otte, Johan? VU University, Center for Advanced Media Research Amsterdam VU University, Department of Artificial Intelligence [email_address] [email_address] An affective virtual agent for natural human-agent interaction Abstract In the present paper we show that an existing computational model of emotion regulation can, if reduced to its reappraisal-specific components, fit skin conductance data obtained from an empirical study of reappraisal. By applying parameter tuning techniques, optimal fits of the model have been found against the (averaged) patterns of the skin conductance data. The errors that were found turned out to be relatively low. Moreover, they have been compared with the errors produced by a baseline variant of the model where the adaptive cycle has been removed, and were found substantially lower. CoMERG: The reappraisal model Results Exp.1 adaptation Exp.1 no adaptation Exp.2 adaptation Exp.2 no adaptation Discussion The model turned out to fit the data quite well. Moreover, the fit of a baseline variant of the model without the adaptation layer was remarkably worse. Background CoMERG produced simulations that match Gross’ theory, but was never validated against empirical data. This paper is the first n c n n v n ERL d v n-norm w n c n Adaptation layer