감정분석 기반의 상담용 생성형 AI모

Abstract

The demand for psychological counseling among students is increasing, but the number of counselors is limited and it is difficult to satisfy the demand. AI-based counseling attempts to solve this shortage, but it is difficult to counsel with directive-based models. There are also models based on deep learning, but they do not consider the connection with the counselor and have the limitation that the entire counseling must be viewed again. In this paper, we propose a generative AI model for counseling based on emotion analysis. This model conducts psychological counseling and emotion analysis, and summarizes the conversation after the end of the counseling, so it is easy to understand the entire counseling. To achieve this goal, the model consists of a Solar-based counseling AI, a summary AI, and an ELECTRA-based analysis AI model. Then, using the collected data, we use various fine-tuning techniques to specialize the models for counseling. In the experiment to validate the model, the analysis AI model showed an F1 Score of about 0.69 on 6,640 sentiment data, and the counseling AI, summary AI model was validated by G-Eval Framework. Thus, we demonstrated the validity of the proposed model