Information Retrieval on HotpotQA (IE, Precision, Recall)
72.9Evidence Index (IE)LLM
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LLMtype=inference-time oracle, metadata=10 Topics2026.06 | 72.9 | 59.33 | 42.7 | |
| MCompassRAGmetadata=10 Topics2026.06 | 70.17 | 56.4 | 40.63 | |
| Meta-Chunkingvariant=PPL2026.06 | 66.77 | 65.23 | 21.4 | |
| LLMtype=inference-time oracle2026.06 | 62.63 | 55.83 | 33.5 | |
| SAKI-RAG2026.06 | 58.73 | 55.6 | 30.03 | |
| Meta-Chunkingvariant=MSP2026.06 | 55.7 | 64.3 | 17.97 | |
| RAPTOR2026.06 | 45.43 | 59.63 | 13.7 | |
| DenseXRetrieval2026.06 | 35.6 | 43.17 | 7.03 |