Skip to main content

Table 9 F1 scores (the highest scores italicized)

From: Developing an online hate classifier for multiple social media platforms

 Simple featuresBOWTF-IDFWord2VecBERTAll featuresa
LR0.0620.7640.7680.8280.8910.892
NB0.1300.5050.6060.6010.8850.868
SVM0.0660.4870.6480.7650.8920.883
XGBoost0.4000.7650.7740.8800.9160.924**
FFNN0.0640.7700.7690.8470.8930.894
KBCn/an/an/an/an/a0.388
BOCn/an/an/an/an/a0.084
  1. ** Significant at p < 0.001 (McNemar’s test comparing predictions from XGBoost-BERT and XGBoost-All
  2. aThe features are concatenated into one big vector for each instance and used as input to the classifiers