Table 10.
The performance of CRF taggers with different word embedding (WE) features. Statistically significant performance increase over using no embeddings is marked with*. Statistically significant performance decrease over using both embeddings is marked with ◊.
| Precision | Recall | F1-measure | |
|---|---|---|---|
| No WE feature | 93.25 | 85.86 | 89.41 ◊ |
| Medical domain WE only | 93.51 | 86.23 | 89.72*◊ |
| Open domain WE only | 93.03 | 86.84 | 89.83*◊ |
| Medical WE + Open WE | 93.22 | 87.16 | 90.09* |