Table 2.
Performance of the best model training, BiLSTMa + ELMo + Flair + semantic embedding, on different sizes of the training corpus.
| Amount of training data (%) | F-1 (%) | Prec (%) | Rec (%) |
| 10 | 71.13 | 69.59 | 72.74 |
| 20 | 82.05 | 81.92 | 82.18 |
| 40 | 85.36 | 85.83 | 84.90 |
| 60 | 86.33 | 86.81 | 85.86 |
| 80 | 86.92 | 87.42 | 86.43 |
aBiLSTM: bidirectional long short-term memory.