Multi-hop Question Answering on MuSiQue (EM, Acc, F1)
29.3Exact Match (EM)HippoRAG2 + CROSSAUG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HippoRAG2 + CROSSAUGFramework=HippoRAG2, Setting=CROSSAUG2026.05 | 29.3 | — | 41.28 | |
| HippoRAG2Framework=HippoRAG2, Setting=Base2026.05 | 27.9 | — | 39.95 | |
| Mujica-MyGOBackbone=Qwen 2.5-7B, Retriever=bge-large-en-v1.5, Domain Setting=Out-of-domain2025.05 | 26.11 | — | 35.92 | |
| IRCoT + IndexRAGCall type=Multi-call, kb=32026.03 | 24.6 | 26.5 | 35 | |
| LinearRAGFramework=LinearRAG, Setting=-2026.05 | 24.1 | — | 37.23 | |
| HippoRAGCall type=Multi-call2026.03 | 23.8 | 25.5 | 34.7 | |
| IterDRAGBackbone=Gemini 1.5, Retriever=Gecko 1B2025.05 | 22.6 | — | 35 | |
| IndexRAGCall type=Single-call, kb=32026.03 | 22.4 | 24.4 | 34.4 | |
| ReSearchBackbone=Qwen 2.5-7B, Retriever=FlashRAG2025.05 | 22.3 | — | — | |
| HippoRAGBackbone=GPT 3.5-turbo, Retriever=ColBERTv22025.05 | 21.9 | — | 33.3 | |
| LightRAG + CROSSAUGFramework=LightRAG, Setting=CROSSAUG2026.05 | 21.4 | — | 29.24 | |
| LightRAGFramework=LightRAG, Setting=Base2026.05 | 20.5 | — | 28.09 | |
| GFM-RAG + CROSSAUGFramework=GFM-RAG, Setting=CROSSAUG2026.05 | 20.4 | — | 30.96 | |
| GFM-RAGFramework=GFM-RAG, Setting=Base2026.05 | 20.3 | — | 30.61 | |
| RAPTORCall type=Single-call2026.03 | 19.3 | 20.3 | 29.7 | |
| Naive RAGCall type=Single-call2026.03 | 19 | 20.4 | 29.9 | |
| FastGraphRAGCall type=Single-call2026.03 | 17.5 | 18.8 | 27.2 | |
| Search-o1Backbone=QwQ 32B, Retriever=Bing Web Search2025.05 | 16.6 | — | 28.2 | |
| Search-R1Backbone=Qwen 2.5-7B, Retriever=E52025.05 | 14.6 | — | — | |
| RAFTBackbone=Llama 3.1-8B, Retriever=Dragon-Plus2025.05 | 13.8 | — | 24 | |
| RAG-StarBackbone=Llama 3.1-8B, Retriever=bge-large-en-v1.52025.05 | 13 | — | 22.2 | |
| RaFeBackbone=GPT 4o-mini, Retriever=Google Search2025.05 | 11.8 | — | 23.8 | |
| IRCoTCall type=Multi-call2026.03 | 11.1 | 12.3 | 19.3 | |
| BM25Call type=Single-call2026.03 | 9.8 | 10.7 | 19.2 | |
| Iter-RetGenBackbone=Qwen 2.5-7B, Retriever=FlashRAG2025.05 | 8.7 | — | — | |
| IRCoTBackbone=Qwen 2.5-7B, Retriever=BM252025.05 | 7 | — | — | |
| RAGBackbone=Qwen 2.5-7B, Retriever=E52025.05 | 5.8 | — | — | |
| Search-o1Backbone=Qwen 2.5-7B, Retriever=E52025.05 | 5.8 | — | — | |
| Mujica (few-shot)Backbone=Qwen 2.5-7B, Retriever=bge-large-en-v1.5, Domain Setting=Out-of-domain2025.05 | 5.34 | — | 17.81 | |
| Direct InferenceBackbone=Qwen 2.5-7B, Retriever=-2025.05 | 3.1 | — | — | |
| CoTBackbone=Qwen 2.5-7B, Retriever=-2025.05 | 2.2 | — | — |