NPOV Rewriting on Wikipedia NPOV (n=107)
68.2VFRTAG
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| TAGExecutor=DeepSeek-V4, #Rules=21.02026.05 | 68.2 | 3.55 | 4.13 | 3.94 | 4.21 | 3.96 | |
| TAGExecutor=Qwen3-30B-A3B-Instruct-2507, #Rules=32.62026.05 | 64.5 | 3.45 | 3.7 | 3.9 | 3.79 | 3.71 | |
| All Rules (M1)Executor=DeepSeek-V4, #Rules=131.02026.05 | 63.5 | 3.51 | 4.24 | 3.96 | 4.21 | 3.98 | |
| All Rules (M1)Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=1312026.05 | 62.6 | 3.33 | 3.99 | 3.92 | 4.33 | 3.89 | |
| Std RAG, top-20Executor=DeepSeek-V4, #Rules=20.02026.05 | 60.8 | 3.48 | 4.44 | 3.84 | 4.29 | 4.01 | |
| Std RAG, top-10Executor=DeepSeek-V4, #Rules=10.02026.05 | 59.8 | 3.34 | 4.38 | 3.76 | 4.08 | 3.89 | |
| Std RAG, top-15Executor=DeepSeek-V4, #Rules=15.02026.05 | 59.8 | 3.39 | 4.4 | 3.79 | 4.3 | 3.97 | |
| Std RAG, top-15Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=152026.05 | 58.9 | 3.31 | 3.79 | 3.96 | 4.35 | 3.85 | |
| Std RAG, top-5Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=52026.05 | 57 | 3.28 | 3.76 | 3.97 | 4.35 | 3.84 | |
| Std RAG, top-5Executor=DeepSeek-V4, #Rules=5.02026.05 | 55.1 | 3.34 | 4.36 | 3.78 | 4.25 | 3.93 | |
| Std RAG, top-10Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=102026.05 | 55.1 | 3.21 | 3.79 | 3.92 | 4.3 | 3.81 | |
| Std RAG, top-20Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=202026.05 | 53.3 | 3.13 | 3.8 | 3.83 | 4.27 | 3.76 | |
| No Retrieval (M0)Executor=DeepSeek-V4, #Rules=0.02026.05 | 32.7 | 2.33 | 4.7 | 3.44 | 4.58 | 3.76 | |
| No Retrieval (M0)Executor=Qwen3-30B-A3B-Instruct-2507, #Rules=02026.05 | 6.5 | 1.25 | 4.94 | 3.13 | 5 | 3.58 |