Table 6. Changes in our model’s performance in DDI detection by removing several features of our model.
| Removed Features | P (%) | R (%) | F (%) | |
|---|---|---|---|---|
| Ensemble | + All Features | 83.6 | 84.0 | 83.8 |
| Single | + All Features | 80.6 | 84.2 | 81.8 |
| Single | + All Features—Static Word Embed | 69.8 | 81.9 | 75.3 |
| Single | + All Features—Subtree feature | 78.6 | 84.2 | 81.2 |
| Single | + All Features—Position feature | 78.0 | 85.5 | 81.4 |
| - Subtree feature | 46.0 | 82.6 | 58.9 | |
| - Static Word Embed | 45.1 | 76.4 | 56.7 |
Note: P, R and F denote Precision, Recall and F1 score, respectively. We test the performance of our single model five times and average the results. For the ensemble performance, we sum the output probabilities of the ensemble members.