Leveraging Social Media and AI for Early Community Mental Health Support

Area of Science:

• Artificial Intelligence in Mental Health
• Computational Psychiatry
• Digital Mental Health Interventions

Background:

• Mental health conditions are a leading cause of global disability.
• Stigma, cost, and access barriers hinder effective mental healthcare.
• Online platforms like Reddit are increasingly used for informal mental health support, but lack professional guidance.

Purpose of the Study:

• To develop and evaluate a framework using AI to detect mental health concerns in social media.
• To generate personalized mental health resources for users based on detected needs.
• To explore AI's potential in complementing traditional mental healthcare delivery.

Main Methods:

• Trained and evaluated machine learning classifiers (Logistic Regression, Random Forest, XGBoost, DistilBERT) for multilabel classification of mental health conditions.
• Utilized Reddit SuicideWatch and Mental Health Collection datasets.
• Employed Llama 3.1 8B Turbo LLM to generate personalized resources based on high-confidence model predictions.

Main Results:

• DistilBERT demonstrated superior performance with an AUC of 0.916, F1 score of 0.762, and accuracy of 0.761.
• The LLM successfully generated tailored mental health resources matched to identified concerns.
• The framework effectively linked symptom detection with resource generation.

Conclusions:

• The AI framework effectively detects mental health concerns and provides personalized resources, addressing barriers to care.
• This approach demonstrates how AI-driven detection can serve as an intervention, guiding users toward support.
• AI has significant potential for scalable, community-based mental health outreach complementing professional clinical care.