Comparative Study for Predicting the Severity of Cyberbullying Across Multiple Social Media Platforms

Initially, cyberbullying detection was mostly done by explicit reporting of inappropriate content by users on the internet, but victims usually are not in the mental space to report such instances, which ultimately go unnoticed by the social media websites. A need for automated methods of cyberbullying detection was realized but the current detection methods require massive improvements to correctly determine the extent of cyberbullying. Our paper aims to identify instances of cyberbullying across various social media platforms and classify them based on the level of severity of cyberbullying. We wish to provide a comparative study of various traditional machine learning and deep learning models that can be used for the above objective.We also wish to release 3 datasets that are annotated in 4 labels based on the severity of cyberbullying (none, low, medium, and high) to be used for future work by the interested person(s). Since young adults are very vulnerable to such incidents, the paper aims to work on social media platforms trending amongst the youth, like Reddit, Twitter, and Formspring.