Offensive Language Detection in Arabic Social Networks Using Evolutionary-Based Classifiers Learned From Fine-Tuned Embeddings

Social networks facilitate communication between people from all over the world. Unfortunately, excessive use of social leads to rise antisocial behaviors such as spread online offensive language, cyberbullying (CB), and hate speech (HS). Therefore, abusive detection become a crucial part cyberharassment. Manual cyberharassment is cumbersome, slow, not even feasible in rapidly growing data. In this study, we addressed challenges automatic tweets Arabic language. The main contribution study design implement an intelligent prediction system encompassing two-stage optimization approach identify classify non-offensive text. first stage, proposed fine-tuned pre-trained word embedding models by training them for several epochs on dataset. embeddings vocabularies new dataset are trained added old embeddings. While second it employed hybrid two classifiers, namely XGBoost SVM, genetic algorithm (GA) mitigate drawback classifiers finding optimal hyperparameter values run approach. We tested Cyberbullying Corpus (ArCybC), which contains collected four Twitter domains: gaming, sports, news, celebrities. ArCybC has categories: sexual, racial, intelligence, appearance. produced superior results, SVM with Aravec SkipGram model achieved accuracy rate 88.2% F1-score 87.8%.