Enhancing Mental Health Detection in Crisis-Related Posts on Social Media: A Domain-Adaptive Transformer Approach
Date:
Abstract
The rise of social media as a platform for expressing mental health concerns, particularly during crises, underscores the need for advanced tools to detect and classify mental health conditions from user-generated text. This study presents MIBERT (Mental Illness BERT), a domain-adaptive transformer model based on BERT, designed to enhance mental health detection in crisis-related social media content.
Methodology
Leveraging a dataset sourced from Reddit subreddits, MIBERT is fine-tuned to classify text across eight categories: ADHD, Anxiety, Bipolar Disorder, Depression, PTSD, OCD, BPD, and None (indicating no mental health condition).
