Conceptual

DiagnosticLlama and Expert-Validated LLM Datasets for PHQ-9 and GAD-7 Symptom Annotation

A fine-tuned Llama model (DiagnosticLlama), obtained by adapting Mentalllama on the PRIMATE dataset, together with expert-clinician-validated ground-truth datasets in which GPT-4o-annotated text spans are mapped to individual PHQ-9 (depression) and GAD-7 (anxiety) diagnostic-criteria symptoms; benchmarked against proprietary and open-source LLMs under naive/exemplar/guidance prompting using hits@k and standard classification metrics, to assist clinicians with structured mental-health diagnostic assessment.