OEmoBERTa: Only Using Emotion for Emotion Recognition in Conversation
Xinye Du · 2022 4th International Conference on Communications, Information System and Computer Engineering (CISCE) · 2022
With the universal application of human-computer interaction systems in people's life, users' requirements for the performance of these systems are gradually improving. For example, when a user uses a smart speaker, they want it to recommend music according to their moods. When treating a patient, if the medical machine can read the current mood of patients and give comfort, it can help doctors complete treatment more efficiently. When users chat with the chatbot, they wish that the robot can feel their emotions and make personalized responses. In order to meet the needs of users and provide them with good services, researchers put a lot of energy on the development of human-computer interaction system. This also makes the number of available public datasets for emotion recognition gradually increase, and more and more scholars put forward their own optimization ideas and developed new models according to the developed models. Thus, in order to make the chatbot respond flexibly according to user's emotion, this research, through improving on the basis of EmoBERTa model, aims to develop a suitable model for this field. Experiments show that the performance of the model is better than other models, the method of reducing input segment makes the prediction time faster than the original model, and the addition of emotion increases the accuracy. The experimental results also confirmed how much the speaker's own emotions can affect the outcome of emotion prediction.