Large language model empowered explainable and interpretable mental health analysis

Area of Science:

• Artificial Intelligence
• Computational Linguistics
• Mental Health Technology

Background:

• Growing concern over mental health expression on social media.
• Limitations of traditional depression detection methods in providing clear explanations.
• Need for interpretable AI in mental health analysis.

Purpose of the Study:

• Introduce InsightDep, an explainable AI model for detecting depression on social media.
• Enhance interpretability of AI-driven mental health analysis using Large Language Models (LLMs).
• Facilitate practical application in therapeutic assessments by medical experts.

Main Methods:

• Utilized a BERT model variant adapted for Twitter data analysis.
• Implemented masked attention techniques for classification and explainability.
• Integrated LLMs to translate complex model explanations into human-understandable formats.

Main Results:

• InsightDep achieved high performance on Twitter and Reddit datasets (macro-F1/accuracy: 0.599/0.671 and 0.994/0.994).
• Outperformed existing state-of-the-art methods like BERTweet, TwHIN-BERT, and DepRoBERTa.
• Demonstrated effective translation of technical results into interpretable insights.

Conclusions:

• InsightDep offers a novel, explainable approach to social media depression detection.
• The methodology supports the development of morally aware digital mental health channels.
• The system is designed for practical use by licensed medical professionals in therapeutic settings.