Identifying schizophrenia stigma on Twitter: a proof of principle model using service user supervised machine...

Area of Science:

• Mental Health
• Computational Linguistics
• Social Media Analysis

Background:

• Stigma surrounding mental health conditions, particularly schizophrenia, discourages help-seeking behaviors.
• Social media platforms like Twitter are significant arenas for public discourse and the formation of attitudes towards mental illness.
• Quantifying public stigma at scale is challenging but crucial for developing targeted interventions.

Purpose of the Study:

• To develop and validate a machine learning pipeline for identifying stigmatizing tweets related to schizophrenia.
• To assess the prevalence of public stigma towards schizophrenia on Twitter.
• To involve service users in the development and evaluation of the machine learning model.

Main Methods:

• A supervised machine learning pipeline was developed with input from a service user group.
• 13,313 public tweets mentioning schizophrenia were collected (January-May 2018).
• A linear Support Vector Machine model, optimized for fewest false negatives, was trained and validated using service user researchers.

Main Results:

• The best-performing model identified public stigma in 47% of the analyzed English tweets.
• Tweets classified as stigmatizing exhibited significantly more negative sentiment compared to non-stigmatizing tweets.
• The machine learning approach, guided by service user input, proved effective in large-scale stigma identification.

Conclusions:

• Machine learning offers a scalable method for detecting and monitoring public stigma on social media.
• The high prevalence of schizophrenia stigma on Twitter underscores the urgent need for public education and online anti-stigma campaigns.
• Machine learning tools can provide real-time metrics to evaluate the effectiveness of anti-stigma initiatives.