POSIGPT: A CHINESE POSITIVE EMOTIONAL TEXT GENERATION MODEL BASED ON PRE-TRAINING METHOD

In recent years, more and more people have been depressed due to the pressure of work and life, and many people have not received enough encouragement and correct guidance in time. Based on this background, this paper proposes a Chinese comment model named PosiGPT. PosiGPT used a generative pre-trained model and used Chinese Weibo data for training. It could detect blogs in depression and generate positive responses. In addition to this model, this paper released a dataset named PosiChat, which contained depression texts and their comments from Sina Weibo. PosiGPT was tested on PosiChat dataset. The results show that the comments generated by PosiGPT have strong fluency, rationality and are always positive, achieving the goal of depression detection and positive response.