Re-ranking on Randomly Retrieved Candidates
0.9167Hit Ratio@3SFT-DPO
Evaluation Results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| SFT-DPOSetting=Supervised Fine-Tuning + Direct Preference Optimization2025.12 | 0.9167 | 0.144 | 0.4802 | 0.5287 | 0.9524 | 0.1893 | 0.3786 | 0.6394 | 1 | 0.331 | 0.331 | 0.4528 | — | — | |
| LLM Re-RankerSetting=Zero-shot2025.12 | 0.8571 | 0.1369 | 0.4563 | 0.4832 | 0.9286 | 0.1964 | 0.3929 | 0.4314 | 0.9762 | 0.2964 | 0.2964 | 0.3503 | — | — | |
| Non-RankerMode=Retriever Baseline2025.12 | 0.7381 | 0.1012 | 0.3373 | 0.3679 | 0.9048 | 0.1714 | 0.3429 | 0.3637 | 1 | 0.3298 | 0.3298 | 0.3468 | — | — |