Improving clinical reliability of LLM reasoning for depression assessment via structured generation and GRPO
Eliseo Bao, Anxo Perez, Javier Parapar · 2026
Read Improving clinical reliability of LLM reasoning for depression assessment via structured generation and GRPO by Eliseo Bao, Anxo Perez, Javier Parapar (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/PMC13400936.txt.
Description
Eliseo Bao, Anxo Perez and Javier Parapar improve, in 2026, how large language models assess depression from social media text, focusing on whether their reasoning can be clinically trusted. Most AI screening tools give a label with little explanation, which limits adoption in care. The team trained models to produce structured rationales aligned with the DSM-5 diagnostic manual.
Models from one to twenty-seven billion parameters were tuned on about 1,500 Reddit posts annotated by a licensed psychologist. A reinforcement learning method rewarding both correct symptom calls and sound reasoning beat standard fine-tuning, with relative gains over ten percent in mid-sized models. Structured rationales added further gains, especially for symptoms needing context, and pushed models to cite relevant evidence.
Catalog record
| CREATOR | Eliseo Bao, Anxo Perez, Javier Parapar |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 59436 |
| LICENSE | https://creativecommons.org/licenses/by/4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |