ED-LLM is a novel multi-task framework that enhances suicide risk detection from social media by combining clinical marker extraction with risk classification.

Key Innovations:

• Uses Mistral-7B in a multi-task learning approach that improves interpretability
• Extracts specific clinical markers from social media text to support risk assessment
• Balances computational efficiency with clinical relevance
• Enables more transparent and explainable risk detection

Medical Impact: This research provides mental health professionals with more reliable, interpretable tools for early intervention, potentially saving lives through timely identification of at-risk individuals on social media platforms.

Evidence-Driven Marker Extraction for Social Media Suicide Risk Detection