Improving Clinical Reliability of LLM Reasoning for Depression Assessment via Structured Generation and GRPO
| UDC.coleccion | Investigación | |
| UDC.departamento | Ciencias da Computación e Tecnoloxías da Información | |
| UDC.grupoInv | Information Retrieval Lab (IRlab) | |
| UDC.institutoCentro | CITIC - Centro de Investigación de Tecnoloxías da Información e da Comunicación | |
| UDC.issue | 3 | |
| UDC.journalTitle | Science Progress | |
| UDC.volume | 109 | |
| dc.contributor.author | Bao, Eliseo | |
| dc.contributor.author | Pérez, Anxo | |
| dc.contributor.author | Parapar, Javier | |
| dc.date.accessioned | 2026-08-17T09:06:30Z | |
| dc.date.available | 2026-08-17T09:06:30Z | |
| dc.date.issued | 2026-07 | |
| dc.description | The ReDSM5 dataset used in this study is publicly available at https://huggingface.co/datasets/irlab-udc/redsm5. Code for model training and evaluation is available at https://github.com/IRLab-UDC/grpo-dsm5/. | |
| dc.description.abstract | [Abstract]: Objective. Digital mental health screening is increasingly explored through the use of AI systems. Yet, most models provide limited insight into how predictions are derived, restricting clinical trust and patient-centered adoption. We investigate whether Large Language Models (LLMs) can generate structured DSM-5-aligned rationales for depression assessment when trained with Group Relative Policy Optimization (GRPO), a reinforcement learning method that encourages outputs aligned with DSM-5 diagnostic criteria. Methods. We fine-tuned LLMs (1B–27B parameters) on the ReDSM5 dataset, covering 1,484 Reddit posts annotated by a licensed psychologist for DSM-5 depressive symptoms and accompanied by expert rationales. We compared standard Supervised Fine-Tuning (SFT) with GRPO-based optimization using a composite reward integrating symptom classification accuracy and the quality of generated reasoning judged against clinical rationales. Results. GRPO consistently improved symptom detection over SFT, with relative gains exceeding 10% for mid-sized models and weighted F1 scores above 0.60. Models trained to generate structured DSM-5-aligned rationales exhibited additional performance boosts (0.09–0.39 F1), particularly for complex symptoms requiring complex contextual interpretation. Qualitative analysis shows that GRPO encourages models to reference symptom-relevant evidence rather than relying on superficial cues. Conclusions. GRPO enables LLMs to produce clinically grounded explanations while improving classification accuracy, representing a promising direction for interpretable AI in social media-based mental health screening, with potential to support patient-centered applications pending clinical validation. | |
| dc.description.sponsorship | The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This workwas supported by the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia (grant ED481A-2024-079); by the Ministry of Science, Innovation and Universities of the Government of Spain (project PID2022-137061OB-C21, MCIN/AEI/10.13039/501100011033); by the Department of Education, Science, Universities, and Vocational Training of the Xunta de Galicia (grant GRC ED431C 2025/49); and by the Xunta de Galicia and the European Union through the FEDER Galicia 2021–2027 OperationalProgramme (Ref. ED431G 2023/01). | |
| dc.description.sponsorship | Xunta de Galicia; ED481A-2024-079 | |
| dc.description.sponsorship | Xunta de Galicia; ED431C 2025/49 | |
| dc.description.sponsorship | Xunta de Galicia; ED431G 2023/01 | |
| dc.identifier.citation | E. Bao, A. Perez, and J. Parapar, "Improving Clinical Reliability of LLM Reasoning for Depression Assessment via Structured Generation and GRPO", Science Progress, Vol. 109, Issue 3, Jul 2026, https://doi.org/10.1177/003685042614674 | |
| dc.identifier.doi | 10.1177/003685042614674 | |
| dc.identifier.issn | 2047-7163 | |
| dc.identifier.uri | https://hdl.handle.net/2183/49037 | |
| dc.language.iso | eng | |
| dc.publisher | Sage | |
| dc.relation.isbasedon | https://huggingface.co/datasets/irlab-udc/redsm5 | |
| dc.relation.isbasedon | https://github.com/IRLab-UDC/grpo-dsm5/ | |
| dc.relation.projectID | info:eu-repo/grantAgreement/AEI/Plan Estatal de Investigación Científica, Técnica y de Innovación 2021-2023/PID2022-137061OB-C21/ES/BUSQUEDA, SELECCION Y ORGANIZACION DE CONTENIDOS PARA NECESIDADES DE INFORMACION RELACIONADAS CON LA SALUD - CONSTRUCCION DE RECURSOS Y PERSONALIZACION | |
| dc.relation.uri | https://doi.org/10.1177/003685042614674 | |
| dc.rights | Attribution-NonCommercial 4.0 International | en |
| dc.rights.accessRights | open access | |
| dc.rights.uri | http://creativecommons.org/licenses/by-nc/4.0/ | |
| dc.subject | Mental health | |
| dc.subject | Depression | |
| dc.subject | Social media | |
| dc.subject | DSM-5 | |
| dc.subject | Reasoning | |
| dc.subject | Explainability | |
| dc.title | Improving Clinical Reliability of LLM Reasoning for Depression Assessment via Structured Generation and GRPO | |
| dc.type | journal article | |
| dc.type.hasVersion | VoR | |
| dspace.entity.type | Publication | |
| relation.isAuthorOfPublication | 99ed6581-6dee-442a-9b37-c35da63bef8a | |
| relation.isAuthorOfPublication | c673c8b1-1afc-48f6-85e9-8f29f9cffb91 | |
| relation.isAuthorOfPublication | fef1a9cb-e346-4e53-9811-192e144f09d0 | |
| relation.isAuthorOfPublication.latestForDiscovery | 99ed6581-6dee-442a-9b37-c35da63bef8a |
Files
Original bundle
1 - 1 of 1
Loading...
- Name:
- Bao_Eliseo_2026_Improving_Clinical_Reliability_of_LLM_Reasoning.pdf
- Size:
- 13.07 MB
- Format:
- Adobe Portable Document Format

