Multi-Hop Question Answering on HotpotQA 300 Questions (EM, F1, Faithfulness)
31.3Exact Match (EM)Baseline RAG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Baseline RAGAvg. Claims Used=19.872026.03 | 31.3 | 45.9 | 63.9 | |
| ECRAvg. Claims Used=19.682026.03 | 29.7 | 45 | 62.6 | |
| Random ControlAvg. Claims Used=19.872026.03 | 20.7 | 30.7 | 42.7 |