Emotion Recognition: An Integration of Different Perspectives

Abstract

Automatic emotion recognition describes the computational task of predicting emotion from various inputs including visual information, speech, and language. This task is rooted in principles from psychology such as the model used to categorize emotions and the definition of what constitutes an emotional expression. In both psychology and computer science, there is a plethora of different perspectives on emotion. The goal of this work is to investigate some of these perspectives about emotion recognition and discuss how these perspectives can be integrated to create better emotion recognition systems. To accomplish this, we first discuss psychological concepts including emotion theories, emotion models, and emotion perception, and how this can be used when creating automatic emotion recognition systems. We also perform emotion recognition on text, visual, and speech data from different datasets to show that emotional information can be expressed in different modalities.