Mental Health Stigma Benchmark Dataset

A theory-grounded, human-annotated benchmark of 470 news and social media excerpts for evaluating the detection and classification of mental health stigma in online communication.

The dataset covers six mental health conditions and includes a fine-grained taxonomy of 28 stigma categories, capturing different modes, domains, and components of stigma. It supports the systematic evaluation of large language models and other automated approaches to stigma detection.

The benchmark includes adjudicated annotations, annotation guidelines, and evaluation resources. Some Reddit text is available to approved researchers upon request.

Paper: Automatic Evaluation of Mental Health Stigma in Online Communication (Baes et al., 2026)

Dataset: GitHub | Hugging Face