Modelling Composite Emotions in Affective Agents

It is important for artificial agents to accurately infer human emotions in order to provide believable interactions. However, there is currently a lack of empirical results supporting affective agent to propose effective computational models for this purpose through analyzing individual profile information and the interaction outcomes. In this paper, we bridge this gap with a game-based empirical study. We propose a general model for interactions between an agent and a user in competitive game settings. Based on results from over 450 players in over 2,500 game sessions, we construct a regression model using a player´s education level, age, gender and the interaction outcome as explanatory factors to compute his/her composite emotions consisting of the six basic emotions.