Detecting and Measuring Depression on Social Media Using a Machine Learning Approach: Systematic Review

Area of Science:

• Computational psychiatry
• Digital mental health
• Natural Language Processing (NLP) in healthcare

Background:

• Depression is a significant global public health concern.
• Early detection of depression is crucial for effective management.
• Social media platforms offer vast text data for mental health research.

Purpose of the Study:

• To systematically review studies applying machine learning (ML) to social media text for depression detection.
• To summarize findings on ML methods and their effectiveness.
• To identify future research directions in this domain.

Main Methods:

• A comprehensive bibliographic search was performed across multiple databases (January 1990-December 2020).
• Two reviewers independently screened 418 identified studies.
• Seventeen studies met the inclusion criteria for the systematic review.

Main Results:

• Machine learning approaches, particularly supervised learning (13/17 studies), were predominantly used.
• Depression identification varied: researcher-inferred mental status (10 studies), user self-descriptions (5 studies), and community membership (2 studies).
• Key challenges include sampling, feature optimization, generalizability, privacy, and ethical considerations.

Conclusions:

• Machine learning applied to social media text shows promise for effective depression detection.
• These ML approaches can serve as valuable complementary tools in public mental health.
• Further research is needed to address identified challenges for broader implementation.