A neuro-fuzzy based intelligent agent for text based emotion recognition

Abstract

Emotion recognition is an important aspect of Affective Computing. This paper deals with the development of an intelligent agent for automatic recognition of emotion from text based events. The approach chosen is Soft Computing and the architecture used is a Neuro-Fuzzy system. The input (event) string is divided into tokens which are then compared to a standard corpus (i.e.,Wordnet-Affect) of emotional keywords. The computed values of emotional weight and polarity are then processed by a Neuro-Fuzzy Controller, which generates the emotion underlying the event. The system considers Ekman's six basic emotions {Happiness, Despair, Disgust, Fear, Anger, Surprise}. The controller is trained to generate correct output through the backpropagation algorithm. The system is implemented using Java, and, Matlab is used for mathematical analysis. The performance of the system is graded based on the measures of Precision and Accuracy. © 2013 Springer.