Mental Disorder Diagnostics with LLMs

The project explores automatic identification and verification of DSM diagnostic criteria within social media text (specifically Reddit posts), utilizing a structured three-step LLM pipeline: (1) categorizing mental health conditions, (2) extracting specific DSM criteria from identified categories, and (3) verifying the accuracy of criterion predictions. The primary objective is to scale up precise synthetic data generation using arbitrary rubrics in a non-verifiable domain with minimal human annotation.