A Systematic Literature Review of Machine Learning Models for Detecting Mental Health Markers in Social Media
Author(s)
Konkolics Possobom, Vinicius
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Social media offers unprecedented opportunities to detect and understand mental health–related phenomena through large-scale, real-time analysis of user-generated content. This thesis presents a systematic literature review of machine learning (ML) approaches for detecting six mental health markers—loneliness, cyberbullying, gratitude, self-criticism, politeness, and social support—in social media data. Each marker was examined through a structured evaluation framework assessing predictive performance, data realism, annotation quality, generalizability, and reusability/interpretability. Results show a clear progression from feature-engineered and recurrent neural network models to transformer-based architectures, with emerging interest in generative and alignment-tuned large language models. While state-of-the-art models achieve strong within-domain performance, cross-platform and cross-lingual generalization remains limited. Reusability and interpretability vary widely, with open, well-documented frameworks concentrated in cyberbullying and politeness detection, while gratitude, loneliness, and social support lag in open resources and transferability. The review identifies methodological gaps, particularly in domain adaptation, annotation consistency, and deployment ethics, and offers practical recommendations for selecting and implementing models suited for large-scale, multi-platform mental health assessment.
Sponsor
Date
Extent
Resource Type
Text
Resource Subtype
Undergraduate Thesis