Multilingual Cyberbullying Detection System

Abstract

As the use of social media has evolved in recent times, so has the ability to cyberbully victims using it. The last decade has witnessed a surge of cyberbullying-this bullying is not only limited to English but also happens in other languages. A large number of mobile device users are in Asian countries such as India. Such a large audience is a fertile ground for cyberbullies -hence, it is very important to detect cyberbullying in multiple languages. Most of the current approaches to identify cyberbullying are focused on English text, and a very few approaches are venturing into other languages. This paper proposes a Multilingual Cyberbullying Detection System for detection of cyberbullying in two Indian languages- Hindi and Marathi. We have developed a prototype that operates across data sets created for these two languages. Using this prototype, we have carried out experiments to detect cyberbullying in these two languages. The results of our experiments show an accuracy up-to 97% and Fl-score up-to 96% on many datasets for both the languages.