Ziyuan Zhuang, Zhiyang Zhang, Sitao Cheng, Fangkai Yang, Jia Liu, Shujian Huang, Qingwei Lin, Saravan Rajmohan, Dongmei Zhang, Qi Zhang. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Abstract This survey aims to investigate, analyze, and compare the state‐of‐the‐art chatbots' feasibility and defects for psychotherapy. The survey points out a series of tasks necessary for future psychotherapy chatbots. We searched about 1200 related literature in public databases and selected five typical and state‐of‐the‐art psychotherapy chatbots. Most of the state‐of‐the‐art psychotherapy chatbots use retrieval‐based methods to generate dialogs. Some psychotherapy chatbots incorporate psychological theories, such as cognitive behavior therapy, to solve unique psychological problems. The assessments show that chatbots can preliminarily recognize specific kinds of negative emotions and give relatively appropriate responses. The randomized controlled trials prove that psychotherapy chatbots are useful for some people with a mental health condition. Compared with real psychologists, psychotherapy chatbots have some advantages, such as accessibility without the limitation on time or location. However, some critical technical obstacles limit the usage of psychotherapy chatbots. The limitations require a series of necessary tasks for more effective and safer psychotherapy chatbots, such as collecting standard, valid, real, and rich corpora. In conclusion, current psychotherapy chatbots can hardly replace human psychologists in the short term, but they can improve human psychologists' effectiveness and efficiency as an auxiliary tool.