dsm-5-tr-f77dbec3·1 events·first seen Aliases: DSM-5-TR
A new arXiv preprint proposes a human-in-the-loop annotation framework for Major Depressive Disorder (MDD) that combines LLM-assisted labeling with expert verification across three stages: evidence selection, DSM-5-TR criterion analysis, and case-level synthesis. A dual-memory architecture (Example Memory and Reflection Memory) internalizes expert feedback to iteratively improve annotations without model retraining. The framework targets dataset construction for explainable AI in mental health rather than clinical diagnosis, and exports reasoning traces and edit histories for auditability. A pilot study reports improved annotation consistency and reduced manual revision effort.