arXiv cs.CL· Olivier Tieleman, Samuel J. Bell, Ziyi Zhu, Ting Su, Elizabeth Kroll, Thomas D. Hull, Caitlin A. Stamatis·· 12 小时前AI 评分42
微调 Qwen 3.5 27B 从 AI 心理健康对话被动估计抑郁严重程度
Fine-tuning LLMs for Passive Depression Severity Estimation from AI Mental Health Dialogue
AI 导读
研究团队微调 Qwen 3.5 27B 并加回归头，仅凭 AI 心理健康应用中的对话预测 PHQ-9 分数，将 3,111 条真实标签与前沿推理模型生成的伪标签扩充至 5,524 名用户。
正文 · 原文
Abstract:Depression is a leading cause of disability worldwide, yet validated questionnaires such as the Patient Health Questionnaire-9 (PHQ-9) often go uncompleted, yielding sparse, biased data. Users of AI mental health applications often describe their mood, sleep and functioning unprompted, producing a continuous record of symptoms. Here, we predict PHQ-9 scores from these conversations alone. As in many clinical language tasks, labels are scarce and imbalanced. We address both by fine-tuning Qwen 3.5 27B with a regression head and augmenting 3,111 ground-truth labels with pseudo-labels from a frontier reasoning model and iteratively trained intermediate models, expanding to 5,524 users with more balanced scores. On 422 held-out users, the model achieved MAE $= 3.98$, Pearson $r = 0.67$, and AUC $= 0.88$ at the clinical threshold (PHQ-9 $\geq 10$); AUC was $> 0.80$ across severity thresholds from PHQ-9 $\geq 3$ to PHQ-9 $\geq 24$, demonstrating that the model captures depression severity across the full clinical spectrum. These results support passive depression monitoring without self-report.
| Comments: | 12 pages, 2 figures |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.17973 [cs.CL] |
| (or arXiv:2606.17973v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2606.17973 arXiv-issued DOI via DataCite |
Submission history
From: Olivier Tieleman [view email]
[v1]
Tue, 16 Jun 2026 14:28:47 UTC (451 KB)
[v2]
Wed, 4 Nov 2026 13:48:53 UTC (304 KB)
来源：arXiv cs.CL · arxiv.org