Knowledge Graph Question Answering on WebQSP
93.3Hit@1RAPL
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RAPLLLM=GPT-4o, Training Strategy=Training with LLM-refined supervision2026.05 | 93.3 | 80.7 | — | — | — | |
| GCRLLM Backbone=ChatGPT2026.01 | 92.6 | 73.2 | — | — | — | |
| GCRLLM Backbone=GPT-4o-mini2026.01 | 92.2 | 74.1 | — | — | — | |
| PATHISELLM=GPT-4.1, Training Strategy=Ours2026.05 | 91.6 | 81.1 | — | — | — | |
| ReGLLM=GPT-4o, Training Strategy=Training with LLM-refined supervision2026.05 | 91.4 | 78.8 | — | — | — | |
| DoGLLM=GPT-4o, Training Strategy=Prompting with in-context learning2026.05 | 91.2 | 56.1 | — | — | — | |
| PATHISELLM=GPT-4o, Training Strategy=Ours2026.05 | 91.2 | 81.3 | — | — | — | |
| SubgraphRAGLLM Backbone=GPT-4o-mini2026.01 | 90.1 | 77.5 | — | — | — | |
| RPO-RAGLLM Backbone=Llama3.1–8B2026.01 | 89.9 | 81.3 | — | — | — | |
| SubgraphRAG (100 triples)LLM=GPT-4o, Training Strategy=Training with weakly supervised paths2026.05 | 89.9 | 76.1 | — | — | — | |
| ORTType=LLM+KGs(non-Fine-tuned)2025.02 | 89.4 | 71.8 | — | — | — | |
| RoECategory=LLMs+KGs, Backbone=Llama3.1-8B-Instruct2025.10 | 89.13 | 74.76 | 1.25 | — | — | |
| DPLLM=GPT-4o, Training Strategy=Training with weakly supervised paths2026.05 | 88.9 | 75.8 | — | — | — | |
| CoGMethod Category=Prompting, Backbone LLM=DeepSeek-V3.22026.06 | 88.7 | — | — | — | — | |
| RPO-RAGLLM Backbone=Llama2–7B2026.01 | 88.3 | 77.8 | — | — | — | |
| DualRLLM=GPT-4, Trainable Retriever GNN=true, Trainable Retriever LLM=true2026.06 | 87.6 | — | — | — | — | |
| RPO-RAGLLM Backbone=Llama3.2–3B2026.01 | 87.3 | 76.4 | — | — | — | |
| PoGLLM=GPT-4, Trainable Retriever LLM=true2026.06 | 87.3 | — | — | — | — | |
| NeuroSymActiveBackbone=LLaMa3-8B2026.02 | 87.1 | — | — | — | — | |
| GCRLLM=GPT-4o, Training Strategy=Training with weakly supervised paths2026.05 | 86.9 | 69.8 | — | — | — | |
| SubGraphRAGCategory=LLMs+KGs, Backbone=Llama3.1-8B-Instruct2025.10 | 86.61 | 70.57 | 3 | — | — | |
| AGE AMARLLM=Llama2-7B-LoRA, Non-parameter Retriever=true2026.06 | 86.5 | — | — | — | — | |
| PathHDType=LLMs + KG2025.12 | 86.2 | 78.6 | — | — | — | |
| PoGLLM=GPT-4o, Training Strategy=Prompting with in-context learning2026.05 | 86.2 | 68.7 | — | — | — | |
| AGE AMARLLM=Llama2-13B-LoRA, Non-parameter Retriever=true2026.06 | 86.2 | — | — | — | — | |
| RoGType=LLM+KGs(Fine-tuned)2025.02 | 85.7 | 70.8 | — | — | — | |
| ROGCategory=Training-based Method2024.03 | 85.7 | — | — | — | — | |
| ROGType=LLMs + KG2025.12 | 85.7 | 70.8 | — | — | — | |
| RoGMethod Category=Fine-tuning2026.06 | 85.7 | — | — | — | — | |
| GNN-RAGLLM=ChatGPT, Trainable Retriever GNN=true2026.06 | 85.7 | — | — | — | — | |
| CoGMethod Category=Prompting, Backbone LLM=GPT-4.1-mini2026.06 | 85.4 | — | — | — | — | |
| GNNRAGCategory=LLMs+GNNs, Backbone=Llama3.1-8B-Instruct2025.10 | 85.25 | 72.01 | 2 | — | — | |
| CRAFTQA2026.06 | 85.2 | — | — | — | — | |
| ReKnoSLLM=GPT-4, Trainable Retriever GNN=true, Trainable Retriever LLM=true2026.06 | 84.9 | — | — | — | — | |
| GoGType=LLMs + KG2025.12 | 84.4 | — | — | — | — | |
| FiDeLiSType=LLMs + KG2025.12 | 84.4 | 78.3 | — | — | — | |
| FiDeLiSLLM Backbone=gpt-4-turbo, Category=Prompting - LLM + KG2024.05 | 84.39 | 78.32 | — | — | — | |
| AMARLLM=Llama2-7B-LoRA, Non-parameter Retriever=true2026.06 | 84.3 | — | — | — | — | |
| PoGMethod Category=Prompting, Backbone LLM=DeepSeek-V3.22026.06 | 83.9 | — | — | — | — | |
| LightPROFBackbone=LLaMa3-8B2026.02 | 83.8 | — | — | — | — | |
| ReKnoSMethod Category=Prompting, Backbone LLM=GPT-4o-mini2026.06 | 83.8 | — | — | — | — | |
| SRPMethod Category=Prompting, Backbone LLM=GPT-4.1-mini2026.06 | 83.6 | — | — | — | — | |
| TrustUQA2026.06 | 83.5 | — | — | — | — | |
| KG-AgentType=LLMs + KG2025.12 | 83.3 | 81 | — | — | — | |
| KG-AgentMethod Category=Fine-tuning2026.06 | 83.3 | — | — | — | — | |
| AMARLLM=Llama2-13B-LoRA, Non-parameter Retriever=true2026.06 | 83.3 | — | — | — | — | |
| KG-HopperLLM=Qwen-2.5-7B, Fine-tuned=true, Training Strategy=Training without intermediate supervision2026.05 | 83.2 | — | — | — | — | |
| RoGCategory=Finetuning - LLM + KG2024.05 | 83.15 | 69.81 | — | — | — | |
| PoGMethod Category=Prompting, Backbone LLM=Qwen3-Coder-30B-A3B2026.06 | 82.9 | — | — | — | — | |
| ToGLLM Backbone=GPT-42026.01 | 82.6 | — | — | — | — | |
| ToGBackbone Model=GPT-42026.05 | 82.6 | — | — | — | — | |
| SGRBackbone Model=GPT-42026.05 | 82.6 | — | — | — | 80.8 | |
| ToGLLM=GPT-4, Trainable Retriever LLM=true2026.06 | 82.6 | — | — | — | — | |
| RPO-RAGLLM Backbone=Llama3.2–1B2026.01 | 82.3 | 69.8 | — | — | — | |
| DeCAFCategory=Finetuning - LLM + KG2024.05 | 82.1 | — | — | — | — | |
| Prior FT SOTA2026.05 | 82.1 | — | — | — | — | |
| DeCAFMethod Category=Fine-tuning2026.06 | 82.1 | — | — | — | — | |
| RoGLLM=LLaMA2-Chat-7B, Fine-tuned=true, Training Strategy=Training with weakly supervised paths2026.05 | 81.9 | 67 | — | — | — | |
| TOGLLM Backbone=gpt-4-turbo, Category=Prompting - LLM + KG2024.05 | 81.84 | 75.97 | — | — | — | |
| Think-on-GraphType=LLMs + KG2025.12 | 81.8 | 76 | — | — | — | |
| GNN-RAGLLM=LLaMA2-Chat-7B, Fine-tuned=true, Training Strategy=Training with weakly supervised paths2026.05 | 81.1 | 69.2 | — | — | — | |
| ToG-2LLM=ChatGPT, Trainable Retriever LLM=true2026.06 | 81.1 | — | — | — | — | |
| ReKnoSLLM=ChatGPT, Trainable Retriever GNN=true, Trainable Retriever LLM=true2026.06 | 81.1 | — | — | — | — | |
| ReadiMethod Category=Prompting, Backbone LLM=GPT-4.1-mini2026.06 | 80.9 | — | — | — | — | |
| AGE G-RetrieverLLM=Llama3.1 8B-LoRA, Non-parameter Retriever=true2026.06 | 80.3 | — | — | — | — | |
| SGRBackbone Model=ChatGPT2026.05 | 80.1 | — | — | — | 78.4 | |
| FiDeLiSLLM Backbone=gpt-3.5-turbo, Category=Prompting - LLM + KG2024.05 | 79.32 | 76.78 | — | — | — | |
| Readi-GPT4Category=Inference-based Method, Backbone=GPT42024.03 | 78.7 | — | — | — | — | |
| RoGCategory=LLMs+KGs, Backbone=Llama3.1-8B-Instruct2025.10 | 78.62 | 64.39 | 5 | — | — | |
| ReasoningLMCategory=Training-based Method2024.03 | 78.5 | — | — | — | — | |
| ToGLLM=GPT-4o, Training Strategy=Prompting with in-context learning2026.05 | 78.5 | 50.9 | — | — | — | |
| DualRLLM=Llama2-13B, Trainable Retriever GNN=true, Trainable Retriever LLM=true2026.06 | 78.3 | — | — | — | — | |
| NuTreaTraining Strategy=Training without intermediate supervision2026.05 | 77.4 | 72.7 | — | — | — | |
| UniKGQAType=Retrieval2025.12 | 77.2 | 72.2 | — | — | — | |
| UniKGQATraining Strategy=Training with weakly supervised paths2026.05 | 77.2 | 72.2 | — | — | — | |
| UniKGQAMethod Category=Fine-tuning2026.06 | 77.2 | — | — | — | — | |
| REAREV2022.10 | 76.4 | 70.9 | — | — | — | |
| ReaRevTraining Strategy=Training without intermediate supervision2026.05 | 76.4 | 70.8 | — | — | — | |
| ToGLLM Backbone=ChatGPT2026.01 | 76.2 | — | — | — | — | |
| ToGBackbone Model=ChatGPT2026.05 | 76.2 | — | — | — | — | |
| SQALERwith_gnn=true2022.10 | 76.1 | — | — | — | — | |
| CoGMethod Category=Prompting, Backbone LLM=Qwen3-Coder-30B-A3B2026.06 | 76 | — | — | — | — | |
| EmQL2022.10 | 75.5 | — | — | — | — | |
| IO PromptingMethod Category=Prompting, Backbone LLM=DeepSeek-V3.22026.06 | 75.3 | — | — | — | — | |
| TIARAMethod Category=Fine-tuning2026.06 | 75.2 | — | — | — | — | |
| TOGLLM Backbone=gpt-3.5-turbo, Category=Prompting - LLM + KG2024.05 | 75.13 | 72.32 | — | — | — | |
| UniKGQACategory=Training-based Method2024.03 | 75.1 | — | — | — | — | |
| UniKGQA2026.02 | 75.1 | — | — | — | — | |
| SGRBackbone Model=Cypher LLM2026.05 | 74.5 | — | — | — | 70.6 | |
| Prior Prompting SOTA2026.05 | 74.4 | — | — | — | — | |
| NSMCategory=Finetuning - LLM + KG2024.05 | 74.31 | — | — | — | — | |
| NSM+hTeacher Network=Hybrid reasoning2021.01 | 74.3 | — | — | — | — | |
| NSMdistilled=true2022.10 | 74.3 | 67.4 | — | — | — | |
| Readi-GPT3.5Category=Inference-based Method, Backbone=GPT3.52024.03 | 74.3 | — | — | — | — | |
| Readi2026.06 | 74.3 | — | — | — | — | |
| RoGLLM=Llama2-7B, Trainable Retriever LLM=true2026.06 | 74.2 | — | — | — | — | |
| G-RetrieverCategory=LLMs+GNNs, Backbone=Llama3.1-8B-Instruct2025.10 | 74.07 | 54.51 | 6.75 | — | — | |
| NSM+pTeacher Network=Parallel reasoning2021.01 | 73.9 | — | — | — | — | |
| KD-CoTCategory=Finetuning - LLM + KG2024.05 | 73.7 | 50.2 | — | — | — | |
| Rigel2022.10 | 73.3 | — | — | — | — |