Table 4.
NER performance of pre-trained language models on EnzChemRED. P, precision; R, recall; F, F1 score.
| PLM | Chemical | Protein | Overall | ||||||
|---|---|---|---|---|---|---|---|---|---|
| P | R | F | P | R | F | P | R | F | |
| AIONER-Bioformer | 86.13 | 85.80 | 85.97 | 76.37 | 78.79 | 77.56 | 82.08 | 82.95 | 82.51 |
| AIONER-PubMedBERT | 87.10 | 85.52 | 86.30 | 79.53 | 74.99 | 77.19 | 84.10 | 81.23 | 82.64 |
| Bioformer + EnzChemRED | 85.83 | 86.87 | 86.35 | 83.13 | 84.75 | 83.93 | 84.73 | 86.01 | 85.36 |
| PubMedBERT + EnzChemRED | 86.92 | 87.21 | 87.07 | 83.19 | 85.07 | 84.11 | 85.38 | 86.33 | 85.86 |
| AIONER-Bioformer + EnzChemRED | 86.20 | 86.95 | 86.58 | 81.90 | 85.35 | 83.59 | 84.41 | 86.30 | 85.35 |
| AIONER-PubMedBERT + EnzChemRED | 87.16 | 87.37 | 87.26 | 83.80 | 86.09 | 84.93 | 85.77 | 86.85 | 86.30 |