Psychiatric symptom identification on D2S (5-fold cross-val)
64.2Macro RecallGPT4o-COT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT4o-COTPrompting Strategy=Chain-of-Thought, Evaluation Protocol=Zero-shot2026.03 | 64.2 | 59.2 | |
| GPT4o-ZSLPrompting Strategy=Direct multi-label classification, Evaluation Protocol=Zero-shot2026.03 | 63.1 | 58.8 | |
| SynSym with RealBackbone=MentalBERT, Training Data=Synthetic + Real (fine-tuned)2026.03 | 58.8 | 61.4 | |
| GPT4o-PSPrompting Strategy=Plan-and-solve, Evaluation Protocol=Zero-shot2026.03 | 57.8 | 57.9 | |
| BERTTraining Data=Real, Evaluation Protocol=Fine-tuned2026.03 | 55.5 | 60 | |
| MentalBERTTraining Data=Real, Evaluation Protocol=Fine-tuned2026.03 | 55.4 | 60.3 | |
| DeBERTaTraining Data=Real, Evaluation Protocol=Fine-tuned2026.03 | 53.5 | 56.3 | |
| SynSym w/o RealBackbone=MentalBERT, Training Data=Synthetic solely, Sample Preprocessing=LLM rewriting for clinical orientation2026.03 | 52.5 | 51.8 |