Text-Based Depression Estimation Using Machine Learning With Standard Labels: Systematic Review and Meta-Analysis.

Area of Science:

• Natural Language Processing
• Machine Learning
• Mental Health

Background:

• Depression significantly impacts daily life and can lead to suicidal behavior.
• Text-based depression estimation offers a feasible approach for early mental health screening.
• Existing reviews often used weak depression labels, limiting model reliability and practical application.

Purpose of the Study:

• Evaluate the predictive performance of text-based depression models using standard labels.
• Identify factors influencing performance heterogeneity, including text resources, representation, model architecture, annotation source, and reporting quality.

Main Methods:

• Systematic literature search following PRISMA 2020 guidelines across four major databases (PubMed, Scopus, IEEE Xplore, Web of Science) from 2014-2025.
• Included studies developed machine learning models using participant-generated text and validated depression labels (clinical diagnosis or scales).
• Conducted random-effects meta-analysis to calculate pooled effect sizes (r) and performed subgroup/meta-regression analyses.

Main Results:

• Analyzed 15 models from 11 studies, revealing a large overall effect size (r=0.605).
• Embedding-based text representations (r=0.741) and deep learning architectures (r=0.731) significantly outperformed traditional features and shallow models, respectively.
• Models using clinician diagnoses (r=0.688) showed higher performance than those using self-report scales (r=0.500); transparent reporting positively correlated with performance (β=0.085).

Conclusions:

• Text-based depression estimation models with standard labels demonstrate robust predictive capabilities.
• Embedding features, deep learning architectures, and clinician diagnoses are key drivers of higher model performance.
• Emphasizes the critical role of standard labels, feature representation, and transparent reporting for enhancing the reliability and practical utility of depression screening models.