Examining the Efficacy of Large Language Models for Mitigating Depression and Anxiety Among Chinese Students: A Randomized Controlled Trial.

Abstract

Secondary school students undergo significant psychological and physiological changes during adolescence, increasing their vulnerability to mental health issues. However, existing mental health services are inadequate to address the growing demand. To bridge this gap, we developed WarmGPT, a conversational mental health service robot utilizing a large language model integrated with cognitive- behavioral therapy, aimed at supporting secondary school students. In this study, 40 students from a Chinese secondary school were randomly assigned to an experimental group or a control group. The experimental group received 2 weeks of counseling through WarmGPT, whereas the control group viewed mental health education videos. Emotional states were evaluated before and after the intervention using scales measuring depression, anxiety, and positive and negative affect. Results indicated that the large language model-based WarmGPT significantly reduced depression, anxiety, and negative emotions and increased positive emotions among the students, outperforming the control group. These findings suggest that large language model-based conversational agents such as WarmGPT are effective in alleviating negative emotions and enhancing overall mental health in secondary school students, offering a promising new approach for mental health interventions.