Complex Emotion Recognition Model based on Conversational Emotion Association

Recently, communication through messengers and SNS(Social Network Service) is increasing due to the spread of non-face-to-face services. Much research is being conducted on user opinion analysis and emotion recognition using the data generated at this time. Conversation data mostly consists of short sentences, limiting the ability to recognize emotions accurately. Additionally, many studies only deal with one representative emotion for one sentence, making it difficult to recognize the emotions inherent in a conversation. In order to recognize accurate emotions in conversation sentences, it is important to derive emotional relationships between conversation sentences and identify multiple inherent emotions. Therefore, in this paper, we classify inherent multi-label emotions by learning emotion classes and conversation sentences through a pre-trained language model optimized for the Korean text processing model. In addition, to understand the combination of emotions according to the flow of conversation sentences in detail, we derive the emotional correlation of conversation sentences and propose a com- plex emotion recognition model that applies the two.