Dynamic Summary Generation for Interpretable Multimodal Depression Detection
Year: 2026
Type: preprint
Abstract: Depression remains widely underdiagnosed and undertreated because stigma and subjective symptom ratings hinder reliable screening. To address this challenge, we propose a coarse-to-fine, multi-stage framework that leverages large language models (LLMs) for accurate and interpretable detection. The pipeline performs binary screening, five-class severity classification, and continuous regression. At each stage, an LLM produces progressively richer clinical summaries that guide a multimodal fusion module integrating text, audio, and video features, yielding predictions with transparent rationale. The system then consolidates all summaries into a concise, human-readable assessment report. Experiments on the E-DAIC and CMDC datasets show significant improvements over state-of-the-art baselines in both accuracy and interpretability.
Source: arXiv (Cornell University)
Authors: Teng, Shiyu, Liu, Jiaqing, Sun, Hao, Li, Yu, Chai, Shurong, Hou, Ruibo, Tateyama, Tomoko, Lin, Lanfen, Chen, Yen-Wei
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7154378293
Cites:
Cited by:
Related to: