Comparative Performance of Machine Learning and Deep Learning Algorithms for Arabic Hate Speech Detection in OSNs

Abstract

Nowadays, Online Social Networks (OSNs) are the most popular and interactive media that used to express feelings, communicate and share information between people. However, along with useful and interesting content, sometimes unsuitable or abusive content can be published on these networks, such as hate speech and insults. Hate speech includes any type of online abuse concepts like cyberbullying, discrimination, abusive language, profanity, flaming, toxicity, and harassment. Most of the Hate speech detection attempts have concentrated on the English text, while work on the Arabic text is sparse. In this paper, we constructed a standard Arabic dataset that can be used for hate speech and abuse detection. In contrast to most previous work the datasets were collected from one platform, the proposed dataset is collected from more social network platforms (Facebook, Twitter, Instagram, and YouTube). To validate the effectiveness of the proposed datasets twelve machine learning algorithms and two deep learning architecture were used. Recurrent Neural Network (RNN) outperformed other classifiers with an accuracy of 98.7%.