Twitter Hate Speech Detection using Stacked Weighted Ensemble (SWE) Model

Online Social Media has expanded the freedom of expression in the internet, which has become a disturbing problem if it has an impact on the situation or the interest of a country. Hate speech refers to the use of hostile, abusive or offensive language, directed at a certain group of people who share common property, whether it is their gender, ethnicity or race (i.e. racism), faith and religion. Therefore, auto detection of hate speeches has an increased importance in Online Social Media for filtering any message that has hatred language before posting it to the network. In this paper, a Stacked Weighted Ensemble (SWE) model is proposed for the detection of hate speeches. The model ensembles five standalone classifiers: Linear Regression, Naïve Bayes', Random Forest, Hard Voting and Soft Voting. The experimental results on a Twitter® dataset has shown an accuracy of 95.54% in binary classification of tweets into hateful speech and an improved performance is noted compared to the standalone classifiers.