Question Answering on NewsQA (test)
73.6F1SpanBERT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SpanBERTTraining protocol=Supervised2020.05 | 73.6 | — | |
| Retro-Readerbackbone=ALBERT2020.01 | 66.8 | 55.9 | |
| DECAPROP2020.01 | 66.3 | 53.1 | |
| ALBERTverifier=TAV2020.01 | 65.9 | 55.3 | |
| Retro-Readerbackbone=ELECTRA2020.01 | 65.7 | 54.7 | |
| ELECTRAverifier=TAV2020.01 | 64.5 | 54 | |
| AMANDA2018.01 | 63.7 | 48.4 | |
| AMANDA2020.01 | 63.7 | 48.4 | |
| NeurQuRI2020.01 | 59.5 | 48.2 | |
| R2-BiLSTM2018.01 | 56.7 | 43.7 | |
| R2-BiLSTM2020.01 | 56.7 | 43.7 | |
| BERT2020.01 | 56.7 | 46.5 | |
| FastQAExtbeam-size=52017.03 | 56.1 | 42.8 | |
| FastQAExt2018.01 | 56.1 | 42.8 | |
| FastQAExtTraining protocol=Supervised2020.05 | 56.1 | 42.8 | |
| FastQAbeam-size=52017.03 | 55.7 | 41.9 | |
| FastQA2018.01 | 55.7 | 41.9 | |
| BiDAF on NewsQA2018.01 | 52.3 | 37.1 | |
| BiDAF2020.01 | 52.3 | 37.1 | |
| Match-LSTM2018.01 | 50 | 34.9 | |
| mLSTMTraining protocol=Supervised, Source=Trischler et al. (2017)2020.05 | 50 | 34.9 | |
| mLSTM2020.01 | 50 | 34.9 | |
| BARB2017.03 | 48.3 | 34.1 | |
| BARB2018.01 | 48.2 | 34.1 | |
| BARB2020.01 | 48.2 | 34.1 | |
| Match-LSTM2017.03 | 48 | 33.4 | |
| AGARBackbone=InternVL3.5-8B2026.06 | 47.2 | — | |
| REFQATraining protocol=Unsupervised, Iterative Data Refinement=true2020.05 | 45.1 | 32.1 | |
| UniformBackbone=InternVL3.5-8B2026.06 | 44.9 | — | |
| RandomBackbone=InternVL3.5-8B2026.06 | 44.6 | — | |
| Plain VQABackbone=InternVL3.5-8B2026.06 | 44.4 | — | |
| VARBackbone=InternVL3.5-8B2026.06 | 42.6 | — | |
| REFQATraining protocol=Unsupervised, Iterative Data Refinement=false2020.05 | 41 | 27.6 | |
| VEA-BrBackbone=InternVL3.5-8B2026.06 | 40.8 | — | |
| AGLABackbone=InternVL3.5-8B2026.06 | 40.6 | — | |
| VEA-ConBackbone=InternVL3.5-8B2026.06 | 40.3 | — | |
| Neural BoW Baseline2017.03 | 36.6 | 24.1 | |
| Neural BoW Baseline2018.01 | 36.6 | 24.1 | |
| CGRBackbone=InternVL3.5-8B2026.06 | 29 | — | |
| Lewis et al. (2019)Training protocol=Unsupervised, Backbone=BERT-Large (Whole Word Masking), Re-implementation=Current Paper2020.05 | 27 | 17.9 |