Harnessing large language models in virtual CBT for university students’ academic anxiety: a preliminary randomized trial

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Academic anxiety is a common mental health problem among university students. It is often associated with reduced learning efficiency and an increased risk of depression. Cognitive behavioral therapy (CBT) is supported by an established evidence base, but its use in university settings is still limited by factors such as therapist availability, time, and physical space. To explore a new approach to digitally assisted intervention, this study integrated virtual reality (VR), large language model (LLM), and retrieval-augmented generation (RAG) technologies. We developed an LLM-VR-CBT system based on CBT principles and preliminarily examined its short-term intervention effects on academic anxiety among university students. This study used a three-arm randomized controlled design. A total of 60 university students with academic anxiety were included and randomly assigned to the LLM-VR-CBT group, traditional CBT group, or minimal-support control group, with 20 participants in each group. The intervention lasted 4 weeks. The linear mixed-effects model results showed significant group-by-time interactions for academic anxiety and heart rate. The LLM-VR-CBT group and traditional CBT group showed significantly greater reductions in academic anxiety and heart rate than the minimal-support control group. The difference in change between the two active intervention groups did not reach statistical significance. Skin temperature did not show a significant group-by-time interaction. These findings suggest that the LLM-VR-CBT system may help reduce academic anxiety among university students in the short term. Because this was a small, short-term exploratory trial and adverse events were not systematically collected as prespecified safety endpoints, future studies with larger samples, multicenter designs, long-term follow-up, and predefined safety monitoring are needed to further evaluate the system’s stability, safety, feasibility, and application boundaries.