Evidence-certified ranking on MAVEN-ERE (test)
71NDCG@10RM-only + Align
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| RM-only + AlignCalls=02026.05 | 71 | 56 | 97 | 78 | 46 | 43 | |
| Qwen3-8B-ECPOCalls=12026.05 | 70 | 55 | 94 | 84 | 65 | 58 | |
| LambdaMART/LTRCalls=02026.05 | 69 | 54 | 96 | 70 | 42 | 40 | |
| Qwen3-8B-GRPOCalls=12026.05 | 65 | 50 | 82 | 64 | 44 | 39 | |
| Qwen3-8B-DPOCalls=12026.05 | 65 | 50 | 83 | 64 | 44 | 39 | |
| GraphRankCalls=02026.05 | 65 | 50 | 94 | 67 | 40 | 38 | |
| Qwen3-8B-SFTCalls=12026.05 | 61 | 46 | 77 | 57 | 36 | 31 | |
| GRaP-style recognizerCalls=02026.05 | 60 | 45 | 94 | 60 | 39 | 35 | |
| Landmark recognizerCalls=02026.05 | 59 | 44 | 93 | 58 | 38 | 34 | |
| LP-RecognizerCalls=02026.05 | 57 | 42 | 92 | 54 | 35 | 31 | |
| Qwen3-8B zero-shotCalls=12026.05 | 52 | 38 | 48 | 34 | 22 | 18 |