Table 4 Performance comparison across variations of our method on development set
From: Biomedical event extraction with a novel combination strategy based on hybrid deep neural networks
CG | PC | MLEE | |||||||
|---|---|---|---|---|---|---|---|---|---|
Methods | Recall | Precision | F1 Scores | Recall | Precision | F1 Scores | Recall | Precision | F1 Scores |
Proposed | 51.29 | 63.34 | 56.68 | 49.32 | 59.90 | 54.10 | 53.53 | 62.34 | 57.60 |
Single-model | 48.34 | 55.73 | 51.77 | 48.43 | 52.32 | 50.30 | 49.96 | 55.64 | 52.65 |
Single-pipeline-model | 47.68 | 54.80 | 51.00 | 47.49 | 52.95 | 50.07 | 48.43 | 53.73 | 50.94 |
Combination-rule-single | 50.50 | 57.46 | 53.75 | 46.92 | 55.94 | 51.04 | 48.09 | 55.28 | 51.43 |
Combination-rule-all | 54.51 | 52.05 | 53.25 | 50.21 | 47.69 | 48.92 | 52.51 | 48.77 | 50.57 |
EE-probability | 49.16 | 64.22 | 55.69 | 46.74 | 60.60 | 52.77 | 48.43 | 63.72 | 55.03 |
Zero-threshold | 50.74 | 62.24 | 55.90 | 48.90 | 57.06 | 52.66 | 53.02 | 62.30 | 57.29 |
Without-CharCNN | 50.57 | 62.18 | 55.78 | 46.08 | 55.08 | 50.18 | 51.57 | 63.52 | 56.93 |