Knowledge Graph Question Answering on GTSQA Ground-truth seed nodes WikiKG2 (test)
92.66HitsULTRAG-OTS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ULTRAG-OTS2026.01 | 92.66 | 91.05 | 89.29 | |
| ULTRAG-OTSReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 90.81 | 89.39 | 87.18 | |
| ULTRAG-OTSReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 85.7 | 83.71 | 81.08 | |
| SubgraphRAGretrieved triples=2002026.01 | 84.34 | 81.66 | 81.62 | |
| GNN-RAG2026.01 | 76.76 | 74.95 | 74.9 | |
| RoG2026.01 | 76.51 | 74.61 | 73.99 | |
| SubgraphRAG (200)Reasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 73.98 | 71.14 | 70.91 | |
| RoGReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 72.63 | 70.73 | 69.81 | |
| RoGReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 72 | 69.97 | 68.88 | |
| SubgraphRAG (200)Reasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 71.82 | 69.37 | 69.01 | |
| GNN-RAGReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 64.98 | 62.89 | 62.7 | |
| ToG2026.01 | 64.73 | 61.99 | 62.06 | |
| GCR2026.01 | 60.83 | 58.41 | 58.15 | |
| GNN-RAGReasoning LLM=GPT-5, PPR subgraph node limit=30,0002026.01 | 51.91 | 49.36 | 49.59 | |
| GPT-52026.01 | 46.36 | 44.4 | 44.16 | |
| GPT-4.12026.01 | 33.72 | 32.63 | 32.36 | |
| GPT-5-mini2026.01 | 33.42 | 31.92 | 31.82 |