Self-Evolving Human-Centered Framework for Explainable Depression Symptom Annotation

arXiv·low signal

Mental-health datasets often assign depression labels without structured evidence, symptom-level justification, or traceable alignment to DSM-5-TR criteria, undermining explainable-AI systems built on them. This self-evolving, human-centered framework ties annotations to DSM-5-TR criteria to improve transparency and downstream interpretability. Relevant to teams building clinical or high-stakes NLP where label provenance and auditability matter more than raw accuracy.