Natural language processing-based model for detecting cyberbullying on Arabic social media

Cyberbullying is a widespread issue on social media sites, offering psychological and emotional risks, especially to youngsters. Many studies have addressed cyberbullying in English, but not in Arabic, especially the Egyptian dialect. According to the study in this paper, NLP approaches were used to detect cyberbullying in Arabic social media content, notably in the Egyptian dialect. A labeled dataset was manually collected from Facebook, Twitter, and Instagram. Several machine learning algorithms have achieved outstanding performance in classifying the data. Most studies classify data into two categories (i.e., coarse-grained) —bullying and non-bullying—but this study divides the dataset into 16 categories, including religious, academic, social, ethnic, political, and others. This fine-grained methodology helps the model locate and analyze social media bullying content. K-Nearest Neighbors had 86% accuracy, Random Forest Classifier 92.7%, Naive Bayes 92.9%, Logistic Regression 93%, and Support Vector Classifier 95.5%, the greatest accuracy. These results show that NLP and machine learning can identify dangerous information, making Arabic-speaking online communities safer.