Arabic Hate Speech Detection Using Deep Recurrent Neural Networks

With the vast number of comments posted daily on social media and other platforms, manually monitoring internet activity for possible national security risks or cyberbullying is an impossible task. However, with recent advances in machine learning (ML), automatic such posts becomes feasible. There still issue privacy internet; however, this study, only technical aspects designing automated system that could monitor detect hate speech Arabic language were targeted, which many companies, as Facebook, Twitter, others, use to prevent cyberbullying. For task, a unique dataset consisting 4203 classified into seven categories, including content against religion, racist content, gender equality, violent offensive insulting/bullying normal positive comments, negative was designed. The extensively preprocessed labeled, its features extracted. In addition, deep recurrent neural networks (RNNs) proposed classification detection speech. RNN architecture, called DRNN-2, consisted 10 layers 32 batch sizes 50 iterations Another model five hidden layers, DRNN-1, used binary classification. Using models, recognition rate 99.73% achieved classification, 95.38% three classes 84.14% comments. This accuracy high complex language, Arabic, different classes. higher than similar methods reported literature, whether three-class seven-class discussed literature review section.