Multi-hop Question Answering on HotpotQA (400 randomly sampled instances)
66Exact Match (EM)HOLMES
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HOLMESBackbone LLM=GPT-42024.10 | 66 | 78 | |
| GraphRAGBackbone LLM=GPT-4o-mini2024.10 | 58.7 | 63.3 | |
| DAVISBackbone LLM=GPT-4o2024.10 | 56.25 | 73.8 | |
| DAVISBackbone LLM=GPT-4-turbo2024.10 | 55.25 | 71 | |
| GraphReaderBackbone LLM=GPT-42024.10 | 55 | 70 | |
| GPT-4oBackbone LLM=GPT-4o2024.10 | 46.3 | 64.1 | |
| GPT-4-turboBackbone LLM=GPT-4-turbo2024.10 | 44.3 | 60.4 |