Hate Speech and Offensive Language Detection from Social Media

Abstract

In recent years, the advent of social media platforms has led users to freely express their opinions on various subjects, including politics, society, health, education, finance, and even business-related issues. However, this widespread usage of social media has also increased the risk of its misuse by some groups resulting in spreading hate speeches or offensive language. This paper investigates machine learning methods for hate speech classification and compared its results with advanced deep learning models to evaluate its efficiency. The data from Twitter, a popular microblogging social media platform for sharing short digital content, was used for experiments in this study. Each tweet was labeled into one of three categories: hate speech, offensive, neutral. Four machine learning methods were investigated: logistic regression (LR), random forest (RF), naive Bayes (NB), and support vector machine (SVM). The results were compared with two deep learning-based models: recurrent neural networks (RNN) and bidirectional encoder representations (BERT). The overall results indicated that both machine learning and deep learning models were effective for hate speech recognition. The highest overall accuracy was obtained using BERT (87.78%), while SVM produced the best (84.66%) among traditional classifiers. © 2021 IEEE.