Multi-hop Question Answering on HotpotQA (EM, F1, Judge, Tokens, Cw)
67.8EMIRCoT + HippoRAG2
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| IRCoT + HippoRAG2Setting=Retrieval2026.05 | 67.8 | 80.7 | 92.4 | 26.82 | 26,370 | |
| GRASPSetting=Retrieval2026.05 | 62.7 | 77.2 | 89.8 | 12.68 | 8,847 | |
| GeARSetting=Retrieval2026.05 | 55.7 | 67.7 | 79.3 | 13.18 | 16,450 | |
| GRASPSetting=Extended Ctx.2026.05 | 54 | 70.4 | 89 | 5.43 | 47,668 | |
| GraphReaderSetting=Extended Ctx.2026.05 | 51.5 | 67.9 | 87 | 12.23 | 76,367 | |
| GraphReaderSetting=Extended Ctx., N=12026.05 | 48.5 | 64 | 82.5 | 7.96 | 56,417 | |
| ReadAgentSetting=Extended Ctx.2026.05 | 44.5 | 61.2 | 82 | 11.53 | 99,831 |