A Review of Light Gradient Boosting Machine Method for Hate Speech Classification on Twitter

Hate speech is a form of verbal communication with the purpose to insult, to provoke, or to incite the victim. With social media such as Twitter, it becomes easier to spread or find hate speech. To reduce online hate speech, we made a classification system to detect hate speech from Twitter. This system uses LightGBM, which was the development of a Gradient Boosting Decision Tree (GBDT) Algorithm. GBDT was often used for this type of classification, but the outcome was less satisfactory. LightGBM uses Gradient-based One Side Sampling (GOSS) and Exclusive Feature Bundling (EFB). With these two techniques implemented in GBDT, this research experiment's accuracy is 86,05% with a total of 1000 data, a 0.175 learning rate, 30% test data, and 70% training data.