CueSupport-MH: An Explainable Dataset and Interpretable Framework for Mental Health Risk Detection and Response Safety Evaluation
CueSupport-MH is introduced, a 6,400-instance benchmark constructed from public Reddit conversations and annotated for four complementary dimensions: risk label, risk evidence span, protective cue, and response-safety label, and a lightweight interpretable framework for joint risk detection and response-safety classifi...