Table 4
Evaluation Results for Stacking with Various Feature-classifier Combinations
| Feature Vector − Classifier Combination | Precision | Recall | F1 Score | AUC |
| metadata(NB) + entity(SVM) + predication(NB) + relation(SVM) + relation(B) | 73.0% | 58.0% | 0.646 | 0.919 |
| word(SVM) + metadata(NB) + entity(NB) + entity(SVM) + entity(B) | 73.7% | 61.5% | 0.670 | 0.892 |
| metadata(NB) + entity(NB) + entity(SVM) + relation(SVM) + relation(B) | 79.6% | 56.5% | 0.661 | 0.912 |
| word(NB) + metadata(B) + entity(B) + predication(SVM) + predication(B) | 26.0% | 88.0% | 0.401 | 0.819 |
| metadata(NB) + entity(NB) + entity(SVM) + predication(SVM) + relation(B) | 72.1% | 62.0% | 0.667 | 0.908 |
-
NB = Naïve Bayes; SVM = Polynomial support vector machine; B = boosting.
