Question Answering on CWQ (Acc, F1)
23.62AccuracyGraph-S^3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Graph-S^3Generator=Finetuned-7B2025.10 | 23.62 | 30.44 | — | |
| G-retrieverGenerator=Finetuned-7B2025.10 | 22.68 | 28.38 | — | |
| RAG/1hopGenerator=Finetuned-7B2025.10 | 19.85 | 26.01 | — | |
| Graph-S^3Generator=Qwen2-7B2025.10 | 17.87 | 23.29 | — | |
| Graph-S^3Generator=Llama3-8B2025.10 | 17.11 | 21.17 | — | |
| LightRAGGenerator=Qwen2-7B2025.10 | 16.2 | 23.09 | — | |
| KG-AgentGenerator=Qwen2-7B2025.10 | 15.64 | 22.45 | — | |
| G-retrieverGenerator=Qwen2-7B2025.10 | 15.38 | 18.62 | — | |
| G-retrieverGenerator=Llama3-8B2025.10 | 13.91 | 17.47 | — | |
| RAG/1hopGenerator=Llama3-8B2025.10 | 13.85 | 17.26 | — | |
| LightRAGGenerator=Finetuned-7B2025.10 | 13.85 | 19.96 | — | |
| KG-AgentGenerator=Finetuned-7B2025.10 | 13.77 | 16.29 | — | |
| RAG/1hopGenerator=Qwen2-7B2025.10 | 12.55 | 16.18 | — | |
| KG-AgentGenerator=Llama3-8B2025.10 | 10.82 | 13.75 | — | |
| No graphGenerator=Finetuned-7B2025.10 | 10.17 | 12.31 | — | |
| No graphGenerator=Llama3-8B2025.10 | 9.4 | 11.58 | — | |
| RAG/2hopGenerator=Finetuned-7B2025.10 | 8.89 | 14.56 | — | |
| ToGGenerator=Llama3-8B2025.10 | 8.42 | 12.33 | — | |
| LightRAGGenerator=Llama3-8B2025.10 | 8.33 | 15.01 | — | |
| ToGGenerator=Finetuned-7B2025.10 | 7.54 | 9.79 | — | |
| ToGGenerator=Qwen2-7B2025.10 | 7.01 | 9.61 | — | |
| RAG/2hopGenerator=Qwen2-7B2025.10 | 7 | 10.55 | — | |
| RAG/2hopGenerator=Llama3-8B2025.10 | 6.29 | 10.68 | — | |
| No graphGenerator=Qwen2-7B2025.10 | 6.26 | 7.35 | — | |
| RAG/3hopGenerator=Qwen2-7B2025.10 | 0.99 | 3.12 | — | |
| RAG/3hopGenerator=Finetuned-7B2025.10 | 0.93 | 4.36 | — | |
| No retrieverGenerator=Finetuned-7B2025.10 | 0.68 | 1.76 | — | |
| RAG/3hopGenerator=Llama3-8B2025.10 | 0.65 | 3.43 | — | |
| No retrieverGenerator=Qwen2-7B2025.10 | 0.37 | 1.29 | — | |
| No retrieverGenerator=Llama3-8B2025.10 | 0.14 | 1.29 | — | |
| ReGradModel=LLaMA-3.2 1B, Top-1000 setting=True2026.06 | — | — | 47.18 | |
| ReGradModel=LLaMA-3.2 3B, Top-1000 setting=True2026.06 | — | — | 52.57 | |
| ReGradModel=LLaMA-3.1 8B, Top-1000 setting=True2026.06 | — | — | 56.41 | |
| ReGrad + ICLModel=LLaMA-3.2 1B, Top-1000 setting=True2026.06 | — | — | 50.34 | |
| ReGrad + ICLModel=LLaMA-3.2 3B, Top-1000 setting=True2026.06 | — | — | 54.08 | |
| ReGrad + ICLModel=LLaMA-3.1 8B, Top-1000 setting=True2026.06 | — | — | 57.52 |