Cyberbullying Detection on Twitter using Multiple Textual Features

Abstract

Due to the spread of PCs and smartphones and the rise of user-generated content in social networking service, cyberbullying is also increasing and has become a serious risk that social media users may encounter. In this paper, we focus on the Japanese text on Twitter and construct an optimal model for automatic detection of cyberbullying by extracting multiple textual features and investigating their effects with multiple machine learning models. The experimental evaluation shows that the best model with predictive textual features is able to obtain an accuracy of over 90%.