This study develops a service user-supervised machine learning pipeline to identify stigmatizing tweets about schizophrenia on Twitter. A service user group guided the model evaluation (prioritizing fewest false negatives) and feature selection. Analyzing 13,313 tweets, a linear Support Vector Machine (SVM) model, chosen for its minimization of false negatives, identified stigma in 47% of English tweets. These tweets exhibited significantly more negative sentiment. The findings highlight the prevalence of schizophrenia stigma online and suggest machine learning's potential for real-time monitoring and evaluation of anti-stigma campaigns.

Sagar Jilka, Clarissa Mary Odoi, Janet van Bilsen, Daniel Morris, Sinan Erturk, Nicholas Cummins, Matteo Cella, Til Wykes