Open-domain Question Answering on NaturalQuestions standard (test)
45.5AccuracyOurs (theory-guided context selection strategy)
Evaluation results
| Method | Links | |
|---|---|---|
| Ours (theory-guided context selection strategy)Model=Qwen3-8B2026.02 | 45.5 | |
| ReMemModel=Qwen3-8B2026.02 | 45.3 | |
| ExpRAGModel=Qwen3-8B2026.02 | 45.1 | |
| DCModel=Qwen3-8B2026.02 | 45 | |
| BM25Model=Qwen3-8B2026.02 | 44.6 | |
| ZeroModel=Qwen3-8B2026.02 | 44.2 | |
| ReMemModel=Llama-3.1-8B2026.02 | 39.8 | |
| Ours (theory-guided context selection strategy)Model=Llama-3.1-8B2026.02 | 39.7 | |
| ExpRAGModel=Llama-3.1-8B2026.02 | 39.4 | |
| DCModel=Llama-3.1-8B2026.02 | 39.3 | |
| BM25Model=Llama-3.1-8B2026.02 | 38.9 | |
| ZeroModel=Llama-3.1-8B2026.02 | 38.5 |