Generative Question Answering on MsMARCO (test)
40.7ROUGE ScoreMatch-LSTM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Match-LSTM2017.03 | 40.7 | 37.3 | |
| FastQAExtk=52017.03 | 33.7 | 33.9 | |
| FastQAk=52017.03 | 32.1 | 34 | |
| ReasoNet2017.03 | 19.2 | 14.8 | |
| PALMModel type=single, Beam search size=5, Batch size=64, Learning rate=1e-5, Max input length=5122020.04 | 0.498 | — | |
| MS-MARCO SotAUses gold context/evidence=true2020.05 | 0.498 | 0.499 | |
| MasqueModel type=ensemble2020.04 | 0.496 | — | |
| BERT+Multi-Pointer2020.04 | 0.495 | — | |
| Selector NLGEN2020.04 | 0.487 | — | |
| VNET2020.04 | 0.484 | — | |
| Communicating BERT2020.04 | 0.483 | — | |
| SNET+CES2S2020.04 | 0.45 | — | |
| KIGN QA2020.04 | 0.441 | — | |
| Reader Writer2020.04 | 0.439 | — | |
| ConZNet2020.04 | 0.421 | — | |
| RAG-SequenceUses gold context/evidence=false2020.05 | 0.408 | 0.442 | |
| RAG-TokenUses gold context/evidence=false2020.05 | 0.401 | 0.415 | |
| BARTUses gold context/evidence=false2020.05 | 0.382 | 0.416 |