Question Answering on NQ Dataset (EM Subtypes)
45.36EM (Long Question)Deep GraphRAG
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Deep GraphRAGKnowledge Integration=Qwen2.5 72B, Generation=Qwen2.5 72B2026.01 | 45.36 | 55.08 | 14.7 | 44.69 | |
| Deep GraphRAGKnowledge Integration=Qwen2.5 72B, Generation=DeepSeek-R12026.01 | 45.36 | 51 | 19.6 | 43.98 | |
| Drift SearchKnowledge Integration=Qwen2.5 72B, Generation=DeepSeek-R12026.01 | 45.23 | 53.42 | 11 | 43.61 | |
| Deep GraphRAGKnowledge Integration=Qwen2.5 1.5B-DW GRPO, Generation=DeepSeek-R12026.01 | 45 | 50.5 | 18 | 42.09 | |
| Deep GraphRAGKnowledge Integration=Qwen2.5 1.5B-DW GRPO, Generation=Qwen2.5 72B2026.01 | 44.8 | 54 | 13 | 42.36 | |
| Drift SearchKnowledge Integration=Qwen2.5 72B, Generation=Qwen2.5 72B2026.01 | 42.43 | 54.15 | 14.7 | 42.78 | |
| Local SearchGeneration=Qwen2.5 72B2026.01 | 41.32 | 16.58 | 16.1 | 30.13 | |
| Local SearchGeneration=DeepSeek-R12026.01 | 41.09 | 18.04 | 23.2 | 31.36 | |
| Deep GraphRAGKnowledge Integration=Qwen2.5 1.5B, Generation=DeepSeek-R12026.01 | 24.4 | 29 | 3.5 | 22.27 | |
| Deep GraphRAGKnowledge Integration=Qwen2.5 1.5B, Generation=Qwen2.5 72B2026.01 | 24 | 28 | 3 | 21.64 | |
| Global SearchKnowledge Integration=DeepSeek-R1, Generation=DeepSeek-R12026.01 | 20.52 | 32.15 | 16.4 | 23.79 | |
| Global SearchKnowledge Integration=Qwen2.5 72B, Generation=Qwen2.5 72B2026.01 | 20.48 | 31.19 | 0.1 | 22.63 |