A Survey of Computational Emotion Research

Abstract

Emotion reserch covers a multi-disciplinary domain. In the past three decades, a number of theoretical models and computer applications have been proposed from different perspectives. There are models of appraisal theory, such as OCC and Frijda’s model, as well as computational systems such as Affective Reasoner, EMA and Cathexis, to make agents believable and human-like. This paper conducts a comprehensive analysis of those state-of-the-art based on the following six criteria, each of which we believe represents a critical aspect in a computational emotion system.

Emotion Type Set. Each emotion model is explicitly or implicitly associated with a set of emotion types, each of which needs a clear characterization. Different models may distinguish different types clearly from others, but it should explain as many as possible, which reflects the adaptability for different emotions.

Emotion Simultaneity. According to psychology research, emotion is not a single and exclusive phenomenon. More than one emotion can arise simultaneously within one emotion experiencer. The ability to handle this feature is important for compatibility.

Role-Orientedness. Human in society have one or more special positions, which are distinguished as social roles. People with different roles may explain and respond to the same event differently, relying on which different emotions arise.

Emotion Situatedness. Most emotions may have an abstract definition or charac-terization. However, such definitions can not be used directly in real-world applications for the gap between abstract characterizations and actual circumstances. A practical system needs the ability to map daily emotional situations into general emotional rules.

Distributed Emotion. People do not live isolated in society. An individual will interact or have various relations with other people at times, thus will influence the emotion response each other by speech and behavior, even by one’s own attitude and emotion state. A broad model may deal with these factors and represent the emotions of the group.

Behavior Display. Following emotions, an individual may express some responses. Moreover, for a given emotional agent in the system, emotions, together with the agent’s goals, may play a key role for the action selection of this agent.