Knowledge Graph Question Answering on WebQSP (Macro-F1 and Hallucination Reporting)
0.9353Hit RateParallaxRAG + GPT-4o (200)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| ParallaxRAG + GPT-4o (200)Generator=GPT-4o, Number of retrieved triples=2002025.10 | 0.9353 | 0.788 | 0.8294 | |
| ParallaxRAG + Qwen3-30B (500)Generator=Qwen3-30B, Number of retrieved triples=5002025.10 | 0.9326 | 0.7611 | 0.7786 | |
| REL-RAG + GPT-4o-miniGenerator=GPT-4o-mini2025.10 | 0.925 | 0.787 | — | |
| ParallaxRAG + Qwen3-30B (200)Generator=Qwen3-30B, Number of retrieved triples=2002025.10 | 0.9248 | 0.7607 | 0.7623 | |
| ParallaxRAG + Qwen3-30BGenerator=Qwen3-30B, Number of retrieved triples=1002025.10 | 0.916 | 0.7524 | 0.7545 | |
| SubgraphRAG + GPT-4oGenerator=GPT-4o2025.10 | 0.898 | 0.7646 | 0.8185 | |
| ParallaxRAG + Llama3.1-8B (↔)Generator=Llama3.1-8B, Cross-dataset generalization=true2025.10 | 0.8951 | 0.6922 | 0.8264 | |
| RoG + GraphRAG-FI2025.10 | 0.8925 | 0.7386 | — | |
| ParallaxRAG + Llama3.1-8BGenerator=Llama3.1-8B, Number of retrieved triples=1002025.10 | 0.8685 | 0.7173 | 0.8364 | |
| RoG-Joint2025.10 | 0.8667 | 0.7026 | 0.7613 | |
| GNN-RAG2025.10 | 0.857 | 0.713 | — | |
| SubgraphRAG + Llama3.1-8B (↔)Generator=Llama3.1-8B, Cross-dataset generalization=true2025.10 | 0.8342 | 0.6642 | 0.8009 | |
| ToG (GPT-4)Generator=GPT-42025.10 | 0.826 | — | — | |
| EtD2025.10 | 0.825 | — | — | |
| RoG-Sep2025.10 | 0.8219 | 0.6645 | 0.7279 | |
| Retrieve-Rewrite-Answer2025.10 | 0.7936 | — | — | |
| StructGPT2025.10 | 0.7469 | — | — | |
| G-Retriever2025.10 | 0.7346 | 0.5341 | 0.6797 | |
| KD-CoT2025.10 | 0.686 | 0.525 | — | |
| SR+NSM w E2E2025.10 | — | 0.641 | 0.6444 | |
| UniKGQA2025.10 | — | 0.722 | — |