Multi-hop Question Answering on MuSiQue (EM, F1, Judge)
46.1Exact Match (EM)Agentic Hybrid
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Agentic HybridCorpus=Medium-scale2026.05 | 46.1 | 57.3 | 66.2 | |
| LOGICALRAGCorpus=Medium-scale2026.05 | 41 | 52 | 59.5 | |
| HippoRAG2Corpus=Medium-scale2026.05 | 36.6 | 48.9 | 51.5 | |
| MA-RAGCorpus=Medium-scale2026.05 | 33.5 | 48.5 | 54.5 | |
| Direct AnswerCorpus=Medium-scale2026.05 | 20 | 32.4 | 34 | |
| CEHop=2-hop, Retrieval Setting=Multi-step2026.07 | 13.34 | — | — | |
| SetCEHop=2-hop, Retrieval Setting=Single-step2026.07 | 11.88 | — | — | |
| BiHop=2-hop, Retrieval Setting=Multi-step2026.07 | 11.72 | — | — | |
| ParaSet+SetCEHop=2-hop, Retrieval Setting=Single-step2026.07 | 10.35 | — | — | |
| SetCEHop=3-hop, Retrieval Setting=Single-step2026.07 | 9.53 | — | — | |
| ParaSet+SetCE (∪ CE)Hop=2-hop, Retrieval Setting=Single-step2026.07 | 9.32 | — | — | |
| ParaSet+SetCE (∪ CE)Hop=3-hop, Retrieval Setting=Single-step2026.07 | 9.32 | — | — | |
| ParaSet+SetCE (∪ CE)Hop=Macro Avg, Retrieval Setting=Single-step2026.07 | 9.3 | — | — | |
| ParaSet+SetCE (∪ CE)Hop=4-hop, Retrieval Setting=Single-step2026.07 | 9.27 | — | — | |
| SetCEHop=Macro Avg, Retrieval Setting=Single-step2026.07 | 8.98 | — | — | |
| CEHop=Macro Avg, Retrieval Setting=Multi-step2026.07 | 8.71 | — | — | |
| BiHop=Macro Avg, Retrieval Setting=Multi-step2026.07 | 8.47 | — | — | |
| BiHop=3-hop, Retrieval Setting=Multi-step2026.07 | 8.19 | — | — | |
| CEHop=3-hop, Retrieval Setting=Multi-step2026.07 | 8.05 | — | — | |
| ParaSet+SetCEHop=Macro Avg, Retrieval Setting=Single-step2026.07 | 7.3 | — | — | |
| ParaSet+SetCEHop=3-hop, Retrieval Setting=Single-step2026.07 | 5.77 | — | — | |
| ParaSet+SetCEHop=4-hop, Retrieval Setting=Single-step2026.07 | 5.77 | — | — | |
| BiHop=4-hop, Retrieval Setting=Multi-step2026.07 | 5.51 | — | — | |
| SetCEHop=4-hop, Retrieval Setting=Single-step2026.07 | 5.51 | — | — | |
| CEHop=4-hop, Retrieval Setting=Multi-step2026.07 | 4.72 | — | — |