Emotion Prediction based on Conversational Context and Commonsense Knowledge Graph

Abstract

Existing studies on emotion recognition in conversations have generally analyzed and classified emotions based on a speaker's utterances from conversations. On the other hand, little research has predicted emotions without such speaker utterances. In this study, we propose an emotion prediction model that forecasts a speaker's emotion before she makes a statement. Our model utilizes the conversational context and commonsense knowledge graphs. In an evaluation experiment, we rate our proposed model’s performance using an emotion recognition dataset in conversations.