Knowledge Graph Question Answering on MetaQA 3-hop
98.9AccuracyReasoning S3GNN-RAG
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Reasoning S3GNN-RAGRAG setting=true2026.05 | 98.9 | — | — | — | |
| Reasoning S3GNNRAG setting=false2026.05 | 98.8 | — | — | — | |
| Reasoning GNNRAG setting=false2026.05 | 98.6 | — | — | — | |
| Reasoning GNN-RAGRAG setting=true2026.05 | 98.6 | — | — | — | |
| Ca2KGBase Model=GPT-3.52026.01 | 89.6 | 5.5 | 6.9 | 95 | |
| Ca2KGBase Model=LLaMA-32026.01 | 87.2 | 6.4 | 7.9 | 95.2 | |
| Verb2S-Top4Base Model=GPT-3.52026.01 | 86.1 | 19 | 10.7 | 92.1 | |
| RoGRAG setting=false2026.05 | 84.8 | — | — | — | |
| Verb1S-Top4Base Model=GPT-3.52026.01 | 83.7 | 9.4 | 12.7 | 88.9 | |
| RC-RAGBase Model=GPT-3.52026.01 | 81.5 | 8.4 | 8.3 | 94.3 | |
| RC-RAGBase Model=LLaMA-32026.01 | 73.6 | 11.8 | 12.1 | 91.3 | |
| Verb2S-Top4Base Model=LLaMA-32026.01 | 73.6 | 11.8 | 12.1 | 91.3 | |
| IoEBase Model=LLaMA-32026.01 | 71 | — | — | — | |
| Verb2S-CoTBase Model=LLaMA-32026.01 | 64.8 | 32 | 28.8 | 75.3 | |
| Verb2S-CoTBase Model=GPT-3.52026.01 | 61 | 52 | 47.1 | 53.3 | |
| Verb1S-Top4Base Model=LLaMA-32026.01 | 59.5 | 9.1 | 8.6 | 94.1 | |
| IoEBase Model=GPT-3.52026.01 | 59 | — | — | — | |
| KG-RAGBase Model=LLaMA-32026.01 | 47 | 24.2 | 26.9 | 69.7 | |
| Self-CorrectBase Model=LLaMA-32026.01 | 34.8 | — | — | — | |
| LLM (No RAG)RAG setting=false2026.05 | 29.7 | — | — | — | |
| Self-CorrectBase Model=GPT-3.52026.01 | 21.9 | — | — | — | |
| KG-RAGBase Model=GPT-3.52026.01 | 17.7 | 82.1 | 81.9 | 11.6 |