Open-domain Question Answering on SQuAD
50.2EMSRC → DS(±)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| SRC → DS(±)Training Strategy=Stage-wise fine-tuning (Source first), Data Augmentation=distant supervision (±)2019.04 | 50.2 | 58.2 | 85.9 | |
| DS(±)Data Augmentation=distant supervision (positive and negative)2019.04 | 48.7 | 56.5 | 85.9 | |
| DS(±) → SRCTraining Strategy=Stage-wise fine-tuning (DS first), Data Augmentation=distant supervision (±)2019.04 | 47.4 | 55 | 85.9 | |
| SRC + DS(±)Training Strategy=Joint (Lumping), Data Augmentation=distant supervision (±)2019.04 | 45.7 | 53.5 | 85.9 | |
| DS(+)Data Augmentation=distant supervision (positive only)2019.04 | 44 | 51.4 | 85.9 | |
| SRCTraining Data=source2019.04 | 41.8 | 49.5 | 85.9 | |
| BERTserini2019.04 | 38.6 | 46.1 | 85.9 | |
| MINIMAL2019.04 | 34.7 | 42.5 | 64 | |
| Par. R.Aggregation Method=Full Agg.2019.04 | 30.2 | — | — | |
| Dr.QAMode=Multitask2019.04 | 29.8 | — | — | |
| Kratzwald and Feuerriegel (2018)2019.04 | 29.8 | — | — | |
| R³2019.04 | 29.1 | 37.5 | — | |
| Par. R.Aggregation Method=Answer Agg.2019.04 | 28.9 | — | — | |
| Par. R.2019.04 | 28.5 | — | 83.1 | |
| Dr.QAMode=Fine-tune2019.04 | 28.4 | — | — | |
| Dr.QA2019.04 | 27.1 | — | 77.8 |