Table 6 Performance comparison with different embedding
Model | macro-P | macro-R | macro-F1 |
|---|---|---|---|
2 layers-Bi-LSTM + word embedding(news corpus) | 83.36% | 83.21% | 83.11% |
2 layers-Bi-LSTM + word embedding(clinical corpus) | 84.55% | 84.53% | 84.49% |
2 layers-Bi-LSTM + character embedding(clinical corpus) | 85.50% | 85.55% | 85.45% |