Review on the Application of Large Language Models in the Mental Health Field

The application of large language models (LLMs) in mental health has become a major direction at the intersection of artificial intelligence and clinical psychology. This review synthesizes recent developments from three perspectives: model characteristics and empirical foundations, clinical applications, and technical advances. At the level of model characteristics and empirical evidence, we examine the inherent properties of LLMs and summarize the empirical support for their use in psychological symptom assessment and mental-health intervention. In terms of applications, we review practical cases and outcomes of LLMs in psychiatric diagnosis, psychological state evaluation, virtual therapy, and clinical decision support. On the technical side, we outline key progress in dataset construction, capability enhancement, and evaluation methodologies tailored to mental-health contexts. Finally, we highlight persistent challenges, including the gap between model outputs and clinical diagnostic practice, limited depth in therapy simulation, scarcity of high-quality datasets, and insufficient clinical validation, and discuss future directions for both clinical deployment and technical research.