Multi-Hop QA on 2Wiki (test)
57.5EMIGMiRAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| IGMiRAG2026.02 | 57.5 | 67.4 | — | |
| Hyper-RAG2026.02 | 45.3 | 54 | — | |
| Naive RAGTop-k retrieved slices=52026.02 | 41.1 | 48.9 | — | |
| NodeRAG2026.02 | 40.9 | 49.7 | — | |
| PathRAG2026.02 | 40.2 | 47.8 | — | |
| Naive RAGTop-k retrieved slices=32026.02 | 37.9 | 45.1 | — | |
| LightRAG2026.02 | 35.2 | 41.4 | — | |
| GPT-4o-mini2026.02 | 32.2 | 37.4 | — | |
| Naive RAGTop-k retrieved slices=12026.02 | 28.8 | 35.1 | — | |
| Cog-RAG2026.02 | 27.5 | 40.7 | — | |
| FLAREType=Agentic RAG2026.04 | — | — | 22.7 | |
| MetaRAGType=Agentic RAG2026.04 | — | — | 29.2 | |
| RAG-CriticType=Agentic RAG2026.04 | — | — | 33.1 | |
| RePAIRTraining Phase=full2026.04 | — | — | 34.5 | |
| RePAIR (offline)Training Phase=offline2026.04 | — | — | 31.5 | |
| RePAIR (online)Training Phase=online2026.04 | — | — | 33 | |
| Self-RAGType=Agentic RAG2026.04 | — | — | 18.9 | |
| Self-RefineType=Agentic RAG2026.04 | — | — | 23.1 | |
| Standard RAGType=Standard RAG2026.04 | — | — | 30.1 |