Attitude Modeling for Virtual Character Based on Temporal Sequence Mining

Virtual agents are increasingly being integrated in our everyday life thanks to their communicative skills and abilities to express social affects like emotions and attitudes. The goal of this work is to evaluate the perception of agents expressing interpersonal attitudes through non-verbal behaviors. The interpretation of these behaviors depends on how they are sequenced and coordinated over time. To encompass the sequentiality and the dynamics of non-verbal signals, we rely on temporal sequence mining. From a multimodal corpus, this algorithm produces meaningful sequences resulting in more adapted expression of social attitudes of the agent.