Cyberbullying Detection and Classification in Social Media Texts Using Machine Learning Techniques

Cyberbullying (CB) is both a public health concern as well as an emotional problem. There have been efforts to mitigate this problem from different discipline dimensions. Artificial Intelligence (AI) has recently emerged as a solution to CB and outshined many earlier solutions. In this work, we have investigated the problem of CB on social media platforms using Machine/Deep Learning and Natural Language Processing (NLP) techniques. Using a dataset containing 47,692 tweets, we investigated the task of detecting CB from social media posts and classifying them as either age-based, religious, ethnic, and political CB or neutral (non-CB). We spot-checked 5 Machine Learning (ML) algorithms (Gradient Boosting, Logistic Regression, Naïve Bayes, Random Forest, and Support Vector Machine) and one Deep Learning algorithm (a sequence model). The algorithms were evaluated based on accuracy, precision, recall, and F1 score. Random Forest reported the best accuracy of 93% while Naive Bayes reported the worst accuracy of 84%, while the DL model had a classification accuracy of 91%. The developed models can help detect and classify CB sentiments in social media posts, thus reducing the harm caused by CB in the social media space.