Multi-hop Question Answering on MuSiQue Ans
34.84Strict Hit RateTGS-RAG
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TGS-RAGBackbone LLM=GPT-4o-mini, Embedding Model=Qwen3-Embedding-0.6B2026.05 | 34.84 | 62.01 | 24.85 | 20.6 | 41.37 | |
| LightRAGRetrieval Mode=Max, Backbone LLM=GPT-4o-mini2026.05 | 33.54 | 60.44 | 20.23 | 18.58 | 40.81 | |
| GraphRAGRetrieval Mode=Local, Backbone LLM=GPT-4o-mini2026.05 | 30.7 | 59.62 | 7.19 | 12.59 | 40.67 | |
| Naive RAGMethod Type=Standard dense retrieval, Backbone LLM=GPT-4o-mini, Embedding Model=Qwen3-Embedding-0.6B2026.05 | 14.23 | 43.96 | 21.17 | 28.11 | 21.51 | |
| Hybrid RAGMethod Type=Dense+Sparse Text Retrieval, Backbone LLM=GPT-4o-mini, Embedding Model=Qwen3-Embedding-0.6B2026.05 | 14.23 | 43.98 | 21.18 | 28.13 | 21.56 | |
| KG2RAGBackbone LLM=GPT-4o-mini2026.05 | 9.1 | 31.02 | 11.5 | 15.61 | 15.06 |