Artificial intelligence in mental health: Knowledge mapping from traditional methods to large language models

Artificial Intelligence (AI) has gained increasing prominence in mental health research over the past decade, offering novel approaches to diagnosis, treatment, and intervention. The rapid emergence of large language models (LLMs) like ChatGPT further reshapes the possibilities within this domain. This study aims to systematically map the evolution of AI research in mental health from traditional machine learning methods to the emergence of LLMs, identifying key research themes, trends, and future directions. A topic modeling analysis was conducted on 4273 peer-reviewed articles published between 2014 and 2025, retrieved from the Web of Science Core Collection. Latent Dirichlet Allocation (LDA) was used to extract major thematic clusters, while keyword co-occurrence and temporal trend analyses were performed to visualize developments over time. Findings revealed a steady rise in publications, with a significant surge post-2020, coinciding with the COVID-19 pandemic. Eight thematic topics were identified, including suicide prediction via social media, computational linguistics, neuroimaging, and digital mental health interventions. Early studies predominantly applied classical algorithms such as support vector machines, transitioning to deep learning and natural language processing. Although LLMs were not yet dominant topic keywords, recent trends indicate their growing presence and transformative potential in mental health applications. This study provides a comprehensive knowledge map of AI in mental health, highlighting its interdisciplinary growth and technological progression. The findings underscore the need for continued integration of advanced AI techniques, particularly LLMs, to support personalized, scalable, and ethically sound mental health care.