Explainable depression symptom detection in social media.

Area of Science:

• Computational linguistics
• Mental health informatics
• Artificial intelligence in healthcare

Background:

• Social media platforms are increasingly used for discussing mental health.
• Existing mental health detection models often lack explainability, hindering clinical trust.
• Clinical markers, like symptoms, are crucial for improving the interpretability of computational models.

Purpose of the Study:

• To develop transformer-based architectures for detecting and explaining depressive symptom markers in social media content.
• To compare separate classification and explanation models with unified models.
• To investigate the utility of conversational Large Language Models (LLMs) for this task.

Main Methods:

• Development of transformer-based models for joint or separate classification and explanation.
• Utilizing in-context learning and fine-tuning with conversational LLMs.
• Evaluation using symptom-focused datasets, offline metrics, and expert-in-the-loop assessments.

Main Results:

• Models successfully detect depressive symptom markers in social media content.
• Generated natural language explanations align with validated clinical symptoms.
• Unified models and LLM approaches show promise in balancing classification accuracy and interpretability.

Conclusions:

• Transformer-based models can effectively detect and explain depressive symptoms on social media.
• Interpretable, symptom-based explanations enhance the utility of computational mental health tools for clinicians.
• This research contributes to building more trustworthy and clinically relevant AI for mental health.