Review of predictive techniques for detecting mental disorders from user-generated content on social media

Mental health status detection via analysis of user-generated content on social media has gained attention. The World Health Organization (WHO) states that one in every eight people in the world lives with a mental disorder. Therefore, mental disorders prediction and prevention are global concerns, and many researchers are exploring the best methods by analysing social media data for it. Existing reviews lack a comprehensive analysis of the techniques, features, and datasets used for mental health status detection using social media data. Hence, this study offers an in-depth review of recent research on mental health status detection on social media platforms, focusing on the predictive techniques employed, features selected, and datasets used. In methodology, a comprehensive searches were conducted across IEEE Xplore, Scopus, ACM Digital Library, ScienceDirect, Wiley Online, SpringerLink, and Google Scholars using the Boolean query: (mental illness OR mental disorder OR mental health status OR mental health state OR mental health analysis) AND (detection OR prediction OR analysis) AND (on OR using OR through) AND (social media OR social media data OR social media platform OR online forums). Out of 1,340 articles published between 2017 and 2024 initially selected, 229 studies met the inclusion criteria following screening and eligibility assessment based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The findings reveal that the most commonly utilised datasets for detecting mental disorders were Twitter (42%), followed by Reddit (12%) and eRisk (10%). Moreover, Textual features, particularly linguistic (67%), were the most commonly used features, followed by emotional features (17%). In addition, Large Language Model (LLM) models like GPT-4 and Llama 3B performed exceptionally well, achieving an accuracy of up to 85% in mental disorders detection tasks. The findings of this review provide valuable insights to researchers, advising on the best predictive techniques, features, and datasets in the field and offering recommendations for future research.