Text-Bullying Detection in Online Social Networks Using Two Levels Ensemble Approach

Increasing use of online social networks and their popularity have provided us with immeasurable options, and the benefits are undeniable. However, people may be embarrassed, insulted, bullied, and harassed by anonymous users, strangers, or peers, degrading these benefits. As a result, there is a significant need to identify, control, and restrict the transmission of bullying information on social networking sites, thus motivating us to perform this study to automate cyberbullying detection. In this paper, we develop a Two Levels Ensemble Detection Bullying (TLEDB) model to detect online text bullying and abusive messages. For this, we select three Machine Learning classifiers and three ensemble models with two feature extraction techniques, namely TF-IDF and Bag of Words (BoW). Experimental results show that our proposed TLEDB model generates the best performance of 98.73% when applying the TF-IDF feature with K-Fold cross-validation and gives an accuracy of 97.55 % when using the BoW feature with Stratified K-fold cross-validation.