Multi-hop Question Answering on MusiqueQA (400 randomly sampled instances)
48EMHOLMES
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HOLMESBackbone LLM=GPT-42024.10 | 48 | 58 | |
| GraphRAGBackbone LLM=GPT-4o-mini2024.10 | 40 | 53.5 | |
| GraphReaderBackbone LLM=GPT-42024.10 | 38 | 47.4 | |
| DAVISBackbone LLM=GPT-4-turbo2024.10 | 34 | 47.1 | |
| DAVISBackbone LLM=GPT-4o2024.10 | 33.8 | 48.5 | |
| GPT-4-turboBackbone LLM=GPT-4-turbo2024.10 | 20.5 | 34.7 | |
| GPT-4oBackbone LLM=GPT-4o2024.10 | 19 | 34.4 |