Semi-Supervised Approach to Monitoring Clinical Depressive Symptoms in Social Media

Area of Science:

• Computational social science
• Digital mental health
• Natural Language Processing (NLP)

Background:

• Social media platforms like Twitter are increasingly used by individuals to express moods and mental health struggles.
• Traditional methods for assessing depression, such as questionnaires, can be intrusive and may not capture real-time emotional states.
• There is a growing need for unobtrusive and scalable methods to screen for mental health conditions.

Purpose of the Study:

• To explore the reliable detection of clinical depression symptoms from social media data (tweets).
• To develop and evaluate a semi-supervised statistical model for identifying depression based on Twitter activity.
• To assess the alignment between expressed symptoms on Twitter and clinical findings from the Patient Health Questionnaire-9 (PHQ-9).

Main Methods:

• Crawling tweets from users who self-reported depressive symptoms.
• Utilizing a semi-supervised statistical model to analyze word usage patterns and topical preferences.
• Evaluating symptom duration and expression in relation to PHQ-9 scores.

Main Results:

• Demonstrated the potential for detecting clinical depression symptoms from tweets, emulating PHQ-9 findings.
• Achieved an accuracy of 68% and a precision of 72% in identifying clinical depressive symptoms.
• Identified correlations between Twitter expression patterns and the severity/duration of depressive symptoms.

Conclusions:

• Social media analysis offers a promising avenue for unobtrusive mental health screening.
• The developed tool can proactively identify potential clinical depression symptoms with significant accuracy.
• This approach can complement traditional diagnostic methods and facilitate early intervention.