Ranking Relevance on Xiaohongshu Search (random)
92.452-ACCProcessRL-Reasoning
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ProcessRL-ReasoningData=150k+50k2025.11 | 92.45 | 81.23 | 73.55 | 80.43 | |
| OutcomeRL-ReasoningData=150k+50k2025.11 | 92.09 | 80.9 | 72.46 | 79.98 | |
| PPO-ReasoningData=150k+50k2025.11 | 91.56 | 78.81 | 70.04 | 79.01 | |
| SFT-LabelData=200k2025.11 | 90.26 | 78.64 | 65.66 | 77.4 | |
| SFT-Reasoning-v2Data=500k2025.11 | 83.04 | 63.06 | 51.76 | 63.16 | |
| SFT-Reasoning-v1Data=150k2025.11 | 80.1 | 59.44 | 47.52 | 59.65 |