Emotion Recognition from Text Using Knowledge-based ANN

This paper proposes an emotion recognition system. Human emotion can be expressed through many kinds of medium such as speech, image, facial expression, and so forth. This paper focuses on the textual data of them. Proposed system is a hybrid system that uses alternatively two methods, keyword-based and machine learning method. Keyword-based methods are traditional approaches using emotional keywords to make a decision of emotional state. They are very correct if emotional keywords exist within text segment. However, there is nothing to do if keywords to be able to catch emotional state do not exist. Therefore, in the case of no emotional keywords, our proposed method uses KBANN to able to infer emotional state through implicit knowledge constructed with the third-party information. Finally, through experiments, we show that the proposed system is more accurate than some previous approaches.