Social media data has become a vital resource for studying mental health, offering real-time insights into thoughts, emotions, and behaviors that traditional methods often miss. Progress in this area has been facilitated by benchmark datasets for mental health analysis; however, most existing benchmarks have become outdated due to limited data availability, inadequate cleaning, and the inherently diverse nature of social media content (e.g., multilingual and harmful material). We present a new benchmark dataset, MindSET, curated from Reddit using self-reported diagnoses to address these limitations. The annotated dataset contains over 13M annotated posts across seven mental health conditions—more than twice the size of previous benchmarks. To ensure data quality, we applied rigorous preprocessing steps, including language filtering, and removal of Not Safe for Work (NSFW) and duplicate content. We further performed a linguistic analysis using LIWC to examine psychological term frequencies across the eight groups represented in the dataset. To demonstrate the dataset’s utility, we conducted binary classification experiments for diagnosis detection using both fine-tuned language models and Bag-of-Words (BoW) features. Models trained on MindSET consistently outperformed those trained on previous benchmarks, achieving up to an 18-point improvement in F1 for Autism detection. Overall, MindSET provides a robust foundation for researchers exploring the intersection of social media and mental health, supporting both early risk detection and deeper analysis of emerging psychological trends.

We introduce ADHD-Lang, a large-scale language resource derived from Reddit to advance computational phenotyping of adult ADHD. The corpus is constructed using a high-precision self-disclosure pattern to confirm ADHD diagnoses and a matched control cohort, comprising 12,070 ADHD users (317,073 posts; 2.83M sentences) and 12,070 controls (174,765 posts; 1.27M sentences). In releasing ADHD-Lang to the research community, we also provide the first comprehensive baseline results, systematically examining the accuracy–transparency trade-off across three model families: (1) interpretable shallow machine learning models trained on clinically meaningful, expert-engineered language biomarkers; (2) a deep BiLSTM network trained on the same feature representations to capture temporal dynamics across users’ posts; and (3) black-box transformer-based models (BERT, RoBERTa, MentalRoBERTa) leveraging contextual embeddings—non-interpretable, high-dimensional representations. ADHD-Lang is released as a standardized benchmark to promote reproducible research and accelerate progress toward digital verbal-behavior phenotyping for adult ADHD.