Multi-hop Question Answering on MuSiQue-Ans (test)
30.4EMRelink
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RelinkMethod Type=Proposed, Backbone LLM=deepseek-v3-03242026.01 | 30.4 | — | — | 41.3 | |
| RAPTORMethod Type=Text-based, Backbone LLM=deepseek-v3-03242026.01 | 25.8 | — | — | 37.1 | |
| HippoRAGMethod Type=Hybrid, Backbone LLM=deepseek-v3-03242026.01 | 25.4 | — | — | 38.1 | |
| GraphRAGMethod Type=Hybrid, Backbone LLM=deepseek-v3-03242026.01 | 20.3 | — | — | 35.8 | |
| G-RetrieverMethod Type=Graph-based, Backbone LLM=deepseek-v3-03242026.01 | 16.6 | — | — | 27.1 | |
| LightRAGMethod Type=Hybrid, Backbone LLM=deepseek-v3-03242026.01 | 12.4 | — | — | 18.3 | |
| Vanilla RAGMethod Type=Text-based, Backbone LLM=deepseek-v3-03242026.01 | 11.8 | — | — | 17.8 | |
| GPT-4oMethod Type=LLM only, Backbone LLM=gpt-4o-2024-07-062026.01 | 10.6 | — | — | 21.3 | |
| ToGMethod Type=Graph-based, Backbone LLM=deepseek-v3-03242026.01 | 9.5 | — | — | 12.7 | |
| Deepseek-v3Method Type=LLM only, Backbone LLM=deepseek-v3-03242026.01 | 6.6 | — | — | 16.7 | |
| Beam Retrievalbeam size=12023.08 | — | 66.9 | 90 | — | |
| Beam Retrievalbeam size=22023.08 | — | 69.2 | 91.4 | — | |
| EE2023.08 | — | 40.7 | 69.4 | — | |
| Ex(EE)2023.08 | — | 46.4 | 78.1 | — | |
| Ex(SA)2023.08 | — | 49 | 80.6 | — | |
| RoHT_mix2023.08 | — | 63.6 | 0 | — | |
| SA2023.08 | — | 52.3 | 75.2 | — |