基於大型語言模型的心理輔助聊天機器人可行性研究

Abstract

憂鬱症是一種常見且嚴重的心理健康問題，影響全球數百萬人。不僅對個人的生活質量產生重大負面影響，對於患者周遭的家人朋友來說是重大的心理負擔，對於社會也是一種負擔。傳統的心理治療方法如認知行為治療(CBT)、焦點解決治療(SFBT)等療法在治療憂鬱症方面已有顯著成效。然而，這些治療方法通常需要由專業的心理治療師進行，而許多患者因資源有限或汙名化而難以獲得適當的治療，也受限於時間空間，難以在需要的時候取得適時的幫助。 近年來，大型語言模型(LLM)的迅速發展為心理健康領域帶來了新的希望。LLM具備處理自然語言的強大能力，能夠生成具有連貫性和上下文相關性的對話。因此，本研究希望探索將LLM應用於心理諮商聊天機器人，建立一個能隨時隨地提供情緒支持的聊天機器人，並在對談過程中融入心理治療療法的方法論，作為一個額外的支持管道，以緩解心理諮商資源不足的困境。 本研究的目的是研究使用大型語言模型開發一個具有使用心理治療療法來與使用者對談的聊天機器人的可行性。聊天機器人會依據患者的憂鬱程度及自殺風險，選擇適當的心理治療療法作為對話的主軸。實驗分為兩個階段進行：第一階段，我們將對話歷史作為輸入，利用LLM預測患者的憂鬱程度及自殺風險，並 依據此資訊對應到不同的心理治療療法。第二階段，我們將為每種心理治療療法設計特定的提示，使LLM能夠根據指定的療法技巧與患者進行對話，提供個性化的治療體驗。在評估階段，我們使用多項指標來衡量聊天機器人的表現，包括對話的相關性、一致性、情緒理解、積極性及技巧正確性。結果顯示，本研究三種心理治療療法的對話在情緒理解以外的四項指標都獲得4分以上或接近4分的評分，顯示我們所設計的framework是有助於讓大型語言模型使用特定心理治療療法來與使用者進行對談的。

Depression is a common and severe mental health issue affecting millions of people worldwide. Besides the impact on individuals' quality of life, it also places a considerable psychological burden on their families, friends as well as the society. Traditional psychotherapies such as Cognitive Behavioral Therapy (CBT), Solution-Focused Brief Therapy (SFBT), etc., have been proved to be effective in depression treatment. However, these treatments often need to be performed by professional therapists, and many patients struggle to access proper treatment due to the limited resource or stigma of mental illness. Recently, the rapid development of large language models (LLMs) has brought new hope to the field of mental health. LLMs have powerful natural language processing capabilities, enabling them to generate coherent and contextually relevant conversations. Therefore, this study aims to explore the application of LLMs in creating a psychological counseling chatbot, designed to provide emotional support anytime and anywhere. This chatbot incorporates the methodologies of psychotherapies into its conversations, serving as an additional support channel to address the shortage of psychological counseling resources. The purpose of this study is to examine the feasibility of developing a chatbot using LLMs that can engage in therapeutic conversations with users. The chatbot will determine the appropriate psychotherapy for conversations based on the user's level of depression and suicide risk. The experiment is conducted in two phases. In the first phase, we use conversation history as input and employ LLMs to predict the user's level of depression and suicide risk. Based on the predicted results, the chatbot determine the psychotherapy it would use as the main structure of the following conversation. In the second phase, a specific set of prompts is designed for each psychotherapy. These prompts will serve as instructions to the LLM, allowing it to engage in a dialogue with the patient based on the specified therapy technique, thereby providing a personalized treatment experience for different users. In the evaluation phase, we use a number of metrics to measure the chatbot's performance, including conversation relevance, consistency, emotional understanding, proactivity, and technical accuracy. The results showed that the conversations of the three psychotherapies in this study all scored above or close to 4 points in all four metrics except emotional understanding, indicating that the framework we designed is helpful for large language models to use specific psychotherapies to talk to users.