Ecmlpkdd2026

Our paper, “Robust Mental Health Detection via Structured LLM Inference: A Study of LangGraph and Chain-of-Verification”, has been accepted in the Research Track of the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD 2026) in September. We study structured inference as a system-level approach to robust mental health detection with LLMs, using a LangGraph-based framework with explicit validation, repair, retry, and abstention to expose failure modes that standard metrics miss. Comparing it against Chain-of-Verification across several models on eRisk anorexia and depression detection tasks, we show that robustness varies substantially across models and cannot be inferred from accuracy alone.