Table 3:
Comparison of single and multi-corpus training on the classification datasets using the RoBERTa-Large transformer model. Multi-corpus training shows performance improvement in datasets annotated with similar guidelines and show minor loss in other datasets.
Training Set | Single-corpus P/R/F1 |
Multi-corpus P/R/F1 |
---|---|---|
DS-NER | 0.97/0.88/0.92 | 0.93/0.92/0.93 |
Tw-NER | 0.76/0.98/0.86 | 0.87/0.86/0.87 |
Micromed | 0.84/0.86/0.85 | 0.83/0.82/0.83 |