Multi-hop Question Answering on Dragonball DragSingleZh (test)
100RecallQwen3-32B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen3-32Bcategory=LLM-in-context, mode=Full-context Inference2026.01 | 100 | 0.01 | |
| Gemini-2.5-Flashcategory=LLM-in-context, mode=Full-context Inference2026.01 | 100 | 0.01 | |
| Gemini-2.5-Procategory=LLM-in-context, mode=Full-context Inference2026.01 | 100 | 0.01 | |
| FABLE(docs)selection_strategy=LLM-guided, bi_path_fusion=true, retrieval_unit=documents2026.01 | 74.99 | 0.4 | |
| FABLE(llm-docs)selection_strategy=LLM-guided only, bi_path_fusion=false, retrieval_unit=documents2026.01 | 74.61 | 5.19 | |
| FABLE(nodes)selection_strategy=LLM-guided, bi_path_fusion=true, retrieval_unit=nodes2026.01 | 72.97 | 3.33 | |
| FABLE(llm-nodes)selection_strategy=LLM-guided only, bi_path_fusion=false, retrieval_unit=nodes2026.01 | 66.07 | 21.9 | |
| TreeRAGcategory=Structure-Enhanced RAG2026.01 | 57.31 | 18.38 | |
| HippoRAG2category=Structure-Enhanced RAG2026.01 | 28.08 | 5.48 | |
| LongRefinercategory=Structure-Enhanced RAG2026.01 | 15.79 | 1.3 |