Table 4.
The impact of pretrained transformer-based models as the backbone for the task of entity pairing and novelty discovering.
| Pretrained model | Entity pair | + Novel |
|---|---|---|
| BioLinkBERT (large) (79) | 75.99 | 53.43 |
| BioGPT (46) | 61.64 | 40.59 |
| BiomedBERT (23) | 72.38 | 49.34 |
Results are presented in terms of F1 score. The best scores are highlighted in bold.