Text-Based Depression Prediction on Social Media Using Machine Learning: Systematic Review and Meta-Analysis.

Area of Science:

• Computational psychiatry
• Digital phenotyping
• Machine learning in mental health

Background:

• Depression impacts over 350 million globally, with traditional diagnostics having limitations.
• Social media text analysis offers novel insights for depression prediction using machine learning.
• A comprehensive review is needed due to the growing research in this field.

Purpose of the Study:

• To evaluate the efficacy of user-generated social media text in predicting depression.
• To assess the impact of demographic, linguistic, social media activity, and temporal features on depression prediction models.
• To synthesize current research on machine learning applications for depression detection via social media.

Main Methods:

• Systematic review of 36 studies published between January 2008 and August 2023.
• Searched 11 major academic databases for relevant research.
• Included studies using social media texts, machine learning, and reporting key performance metrics (AUC, r, sensitivity/specificity).
• Random effects model used for meta-analysis; heterogeneity assessed via forest plots and Cochran Q test.

Main Results:

• A significant large effect size (r=0.630) was found for the correlation between social media texts and depression.
• Demographic features showed the largest effect size (r=0.642), followed by social media activity (r=0.552), language (r=0.545), and temporal features (r=0.531).
• Social media platform type, machine learning approach (shallow vs. deep), and outcome measure selection were significant moderators.

Conclusions:

• Social media text analysis is a promising tool for depression prediction.
• Demographic, linguistic, activity, and temporal features are crucial for enhancing prediction accuracy.
• Further research should consider platform type, ML approach, and outcome measures for robust depression prediction models.