Retrieval-Augmented Generation on 2WikiMultiHopQA
65.04F1 ScoreGraph-R1 (ours)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Graph-R1 (ours)Backbone=Qwen2.5-7B-Instruct, Knowledge Representation=Graph-based, Learning Strategy=Training2025.07 | 65.04 | 82.42 | |
| Graph-R1 (ours)Backbone=Qwen2.5-3B-Instruct, Knowledge Representation=Graph-based, Learning Strategy=Training2025.07 | 57.56 | 76.45 | |
| Search-R1Backbone=Qwen2.5-7B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 41.29 | 70.26 | |
| Search-R1Backbone=Qwen2.5-3B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 38.04 | 54.39 | |
| Graph-R1 (ours)Backbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Graph-based, Learning Strategy=Training2025.07 | 35.13 | 65.73 | |
| R1-SearcherBackbone=Qwen2.5-7B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 33.96 | 69.61 | |
| R1Backbone=Qwen2.5-7B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 30.99 | 59.19 | |
| R1Backbone=Qwen2.5-3B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 28.45 | 56.92 | |
| Search-R1Backbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 28.43 | 60.61 | |
| R1-SearcherBackbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 28.01 | 58.81 | |
| R1Backbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 26.28 | 47.48 | |
| R1-SearcherBackbone=Qwen2.5-3B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 23.5 | 55.86 | |
| StandardRAGBackbone=GPT-4o-mini, Knowledge Representation=Chunk-based, Learning Strategy=Prompting2025.07 | 22.31 | 73.02 | |
| HyperGraphRAGBackbone=GPT-4o-mini, Knowledge Representation=Graph-based, Learning Strategy=Prompting2025.07 | 21.14 | 76.76 | |
| SFTBackbone=Qwen2.5-7B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 20.28 | 63.85 | |
| NaiveGenerationBackbone=GPT-4o-mini, Knowledge Interaction=No, Learning Strategy=Prompting2025.07 | 17.03 | 74.86 | |
| LightRAGBackbone=GPT-4o-mini, Knowledge Representation=Graph-based, Learning Strategy=Prompting2025.07 | 16.59 | 71.94 | |
| HippoRAG2Backbone=GPT-4o-mini, Knowledge Representation=Graph-based, Learning Strategy=Prompting2025.07 | 16.27 | 68.78 | |
| GraphRAGBackbone=GPT-4o-mini, Knowledge Representation=Graph-based, Learning Strategy=Prompting2025.07 | 16.02 | 72.81 | |
| SFTBackbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 13.26 | 34.72 | |
| StandardRAGBackbone=Qwen2.5-7B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Prompting2025.07 | 12.75 | 60.06 | |
| StandardRAGBackbone=Qwen2.5-3B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Prompting2025.07 | 12.52 | 60.01 | |
| PathRAGBackbone=GPT-4o-mini, Knowledge Representation=Graph-based, Learning Strategy=Prompting2025.07 | 12.42 | 67.19 | |
| SFTBackbone=Qwen2.5-3B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Training2025.07 | 12.4 | 52.31 | |
| NaiveGenerationBackbone=Qwen2.5-7B-Instruct, Knowledge Interaction=No, Learning Strategy=Prompting2025.07 | 12.25 | 66.75 | |
| StandardRAGBackbone=Qwen2.5-1.5B-Instruct, Knowledge Representation=Chunk-based, Learning Strategy=Prompting2025.07 | 11.46 | 55.38 | |
| NaiveGenerationBackbone=Qwen2.5-1.5B-Instruct, Knowledge Interaction=No, Learning Strategy=Prompting2025.07 | 7.78 | 49.13 | |
| NaiveGenerationBackbone=Qwen2.5-3B-Instruct, Knowledge Interaction=No, Learning Strategy=Prompting2025.07 | 7.59 | 55 |