Predicting future mental illness from social media: A big-data approach.

Area of Science:

• Computational linguistics
• Psychiatry
• Social media analytics

Background:

• Mental illness diagnosis often relies on self-reporting or clinical observation.
• Early detection of mental health conditions is crucial for timely intervention.

Purpose of the Study:

• To investigate if everyday language in online discussions can predict future mental illness.
• To explore the predictive power of both clinical and nonclinical language samples.

Main Methods:

• Collected language data from Reddit discussion groups (clinical and nonclinical subreddits).
• Utilized natural language processing (NLP) models to analyze word usage.
• Trained models to distinguish mental illnesses and predict future postings to clinical subreddits.

Main Results:

• Words from clinical subreddits distinguished specific mental illnesses (ADHD, anxiety, bipolar disorder, depression).
• Words from nonclinical subreddits also differentiated mental illness categories.
• Nonclinical language predicted future engagement with clinical mental health discussions, indicating early signals.

Conclusions:

• Everyday language contains predictive signals for future mental illness.
• The impact of mental illness extends to non-health-related topics.
• Predictive features for mental illness may evolve, shifting from symptoms to life stressors over time.