Table 4.
The performance of the C-BiLSTM-CRF model (run 1) for each entity type on the test set under the lenient evaluation mode
| Entity Type | P | R | F |
|---|---|---|---|
| Drug | 0.9306 | 0.8742 | 0.931 |
| Strength | 0.9558 | 0.9452 | 0.9741 |
| Duration | 0.7767 | 0.6534 | 0.8161 |
| Route | 0.9284 | 0.9337 | 0.947 |
| Form | 0.9249 | 0.9153 | 0.9494 |
| ADE | 0.4328 | 0.2784 | 0.3875 |
| Dosage | 0.8969 | 0.8829 | 0.9267 |
| Reason | 0.5849 | 0.521 | 0.6267 |
| Frequency | 0.8479 | 0.828 | 0.9695 |
ADE: adverse drug event; C-BiLSTM-CRF: convolutional neural network bidirectional long short-term memory conditional random field; F: micro F-score; P: precision; R: recall.