Multi-hop Question Answering on MuSiQue (Str-Acc, LLM-Acc)
48.4String AccuracyConRAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ConRAGLLM backbone=Gemma-4-31B, Type=Ours2026.05 | 48.4 | 52.3 | |
| Youtu-GraphRAGLLM backbone=Gemma-4-31B, Type=Graph-based RAG2026.05 | 43.8 | 48.3 | |
| ConRAGLLM backbone=GPT-4o-mini, Type=Ours2026.05 | 40.6 | 41.8 | |
| MemGraphRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 34.4 | 37.9 | |
| LinearRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 33.2 | 37.2 | |
| GFM-RAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 32.5 | 36.1 | |
| HippoRAG2Retrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 32.2 | 38.3 | |
| LogicRAGLLM backbone=Gemma-4-31B, Type=Reasoning-based RAG2026.05 | 32.1 | 37.2 | |
| LogicRAGLLM backbone=GPT-4o-mini, Type=Reasoning-based RAG2026.05 | 30.4 | 37.5 | |
| HippoRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 30.4 | 26 | |
| LogicRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 30.1 | 34.6 | |
| LightRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 28.6 | 30.5 | |
| KGPRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 28.4 | 32.5 | |
| RetrievalRetrieval Strategy=Vanilla Retrieval-Augmented-Generation, Top-k Retrieval=52026.05 | 28.3 | 32 | |
| HippoRAG2LLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 27 | 32.6 | |
| RetrievalRetrieval Strategy=Vanilla Retrieval-Augmented-Generation, Top-k Retrieval=32026.05 | 26.6 | 28.5 | |
| E^2GraphRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 26.1 | 29 | |
| HippoRAG2LLM backbone=Gemma-4-31B, Type=Graph-based RAG2026.05 | 25.7 | 27.9 | |
| HippoRAGLLM backbone=Gemma-4-31B, Type=Graph-based RAG2026.05 | 25.4 | 27.6 | |
| RAPTORLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 25.2 | 29.1 | |
| Vanilla RAG (Top-5)LLM backbone=Gemma-4-31B, Type=Text RAG2026.05 | 25.2 | 27.4 | |
| HippoRAGLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 24.9 | 30.1 | |
| RAPTORRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 24.7 | 28.9 | |
| KGPLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 23.3 | 27.5 | |
| Vanilla RAG (Top-3)LLM backbone=Gemma-4-31B, Type=Text RAG2026.05 | 22 | 22.7 | |
| LazyGraphRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 21.5 | 24.8 | |
| Vanilla RAG (Top-5)LLM backbone=GPT-4o-mini, Type=Text RAG2026.05 | 21 | 23.6 | |
| MS-GraphRAGRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 20.6 | 23.7 | |
| Vanilla RAG (Top-3)LLM backbone=GPT-4o-mini, Type=Text RAG2026.05 | 20.3 | 23.6 | |
| RetrievalRetrieval Strategy=Vanilla Retrieval-Augmented-Generation, Top-k Retrieval=12026.05 | 19.9 | 23.8 | |
| LinearRAGLLM backbone=Gemma-4-31B, Type=Graph-based RAG2026.05 | 18.3 | 19.7 | |
| LightRAGLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 18.1 | 19.4 | |
| Zero-shotLLM backbone=GPT-4o-mini, Type=Direct LLM2026.05 | 17.6 | 14 | |
| LinearRAGLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 17.4 | 17.9 | |
| GraphRAGLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 16.5 | 23.1 | |
| G-retrieverRetrieval Strategy=Graph-based Retrieval-Augmented-Generation2026.05 | 16.2 | 17.6 | |
| GPT-4o-miniRetrieval Strategy=Direct Zero-shot LLM Inference2026.05 | 14.1 | 15.2 | |
| Vanilla RAG (Top-1)LLM backbone=Gemma-4-31B, Type=Text RAG2026.05 | 14 | 14.7 | |
| Vanilla RAG (Top-1)LLM backbone=GPT-4o-mini, Type=Text RAG2026.05 | 13.2 | 18.5 | |
| GPT-3.5-TurboRetrieval Strategy=Direct Zero-shot LLM Inference2026.05 | 10.1 | 21.4 | |
| Zero-shotLLM backbone=Gemma-4-31B, Type=Direct LLM2026.05 | 9.3 | 9.3 | |
| G-RetrieverLLM backbone=GPT-4o-mini, Type=Graph-based RAG2026.05 | 9.1 | 15.6 | |
| Llama3-8BRetrieval Strategy=Direct Zero-shot LLM Inference2026.05 | 6.7 | 7.4 | |
| Llama3-13BRetrieval Strategy=Direct Zero-shot LLM Inference2026.05 | 4.2 | 5 |