Entity Linking on AIDA testa CoNLL03 (val)
89.4Strong F1Kolitsas et al. (2018)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Kolitsas et al. (2018)Description=Joint modelling of MD and ED2020.03 | 89.4 | 89.8 | 93.7 | |
| BERT+EntitySetting=Setting II, Training=Both BERT and entity classifier trained on Wikipedia with downsampling2020.03 | 86 | 87.3 | 92.3 | |
| BERT+EntitySetting=Setting I, Training=Both BERT and entity classifier trained on Wikipedia2020.03 | 82.8 | 84.4 | 86.6 | |
| Kolitsas et al. (2018) indep. baselineDescription=Independent modelling of MD and ED, Mention Detection=Stanford NLP2020.03 | 80.3 | 80.5 | — | |
| Frozen-BERT+EntitySetting=Setting I, Training=Frozen BERT, entity classifier trained on Wikipedia2020.03 | 76.8 | 79.6 | 80.6 | |
| Frozen-BERT+EntitySetting=Setting II, Training=Frozen BERT, entity classifier trained on Wikipedia with downsampling2020.03 | 76.5 | 80.1 | 79.6 | |
| BERTTraining=Finetuned on CoNLL03/AIDA, Classifier=Trained from scratch2020.03 | 63.3 | 66.6 | 67.6 |