Mental Health Risk Detection From Social Media Text Data: A Scoping Review of the Machine Learning Research Landscape

Machine learning approaches have been increasingly applied to social media text data for mental health risk detection. However, existing studies vary widely in target outcomes, data sources, labeling strategies, and evaluation practices, and a structured overview of recent research remains limited. This scoping review aims to map the recent research landscape of machine learning-based mental health risk detection using social media text data. Following PRISMA ScR guidelines, peer reviewed journal articles published between January 2021 and January 2026 were retrieved from PubMed, Web of Science, and IEEE Xplore. Studies applying machine learning or deep learning methods to social media text data for mental health risk detection were included and synthesized descriptively. A total of 136 studies were identified. Most focused on depression, anxiety, and suicide or self-harm related risks. Mental health risk was predominantly operationalized through proxy indicators derived from user-generated content, with limited use of survey-linked or clinically anchored labels. Traditional machine learning, deep learning, and Transformer-based models coexisted, alongside substantial heterogeneity in validation strategies and performance metrics. Current research primarily targets proxy-based mental health risk signals rather than clinical diagnoses. This review clarifies prevailing research emphases and methodological practices, and supports the use of social media-based approaches for population-level monitoring and early risk identification.