Conversational Question Answering on CoQA (dev)
0.849Overall F1UNILM
Evaluation Results
| Method | Links | |
|---|---|---|
| UNILMMode=fine-tuned, Epochs=2, Batch size=16, Maximum length=5122019.05 | 0.849 | |
| ERNIE-GENLARGEbeam size=32020.01 | 0.845 | |
| BERT_LARGEModel type=cased, Mode=fine-tuned, Epochs=2, Batch size=16, Maximum length=5122019.05 | 0.827 | |
| UNILMLARGE2020.01 | 0.825 | |
| UNILMfine-tuning epochs=10, batch size=32, mask probability=0.5, maximum length=512, label smoothing rate=0.1, beam size=32019.05 | 0.825 | |
| BiDAF++Context window=3-ctx2018.09 | 0.692 | |
| DrQA+ELMoArchitecture=LSTM-based, Augmentation=pre-trained ELMo representation2019.05 | 0.672 | |
| DrQA + PGNetType=Abstractive2018.09 | 0.662 | |
| BiDAF++Context window=0-ctx2018.09 | 0.634 | |
| DrQAType=Extractive2018.09 | 0.547 | |
| PGNet2020.01 | 0.454 | |
| PGNetarchitecture=Seq2Seq with a copy mechanism2019.05 | 0.454 | |
| Seq2Seq2020.01 | 0.275 | |
| Seq2Seqarchitecture=sequence-to-sequence model with an attention mechanism2019.05 | 0.275 |