Cyberbullying-related Hate Speech Detection Using Shallow-to-deep Learning

Communication in society had developed within cultural and geographical boundaries prior to the invention of digital technology. The latest advancements in communication technology have significantly surpassed the conventional constraints for communication with regards to time and location. These new platforms have ushered in a new age of user-generated content, online chats, social network and comprehensive data on individual behavior. However, the abuse of communication software such as social media websites, online communities, and chats has resulted in a new kind of online hostility and aggressive actions. Due to widespread use of the social networking platforms and technological gadgets, conventional bullying has migrated from physical form to online, where it is termed as Cyberbullying. However, recently the digital technologies as machine learning and deep learning have been showing their efficiency in identifying linguistic patterns used by cyberbullies and cyberbullying detection problem. In this research paper, we aimed to evaluate shallow machine learning and deep learning methods in cyberbullying detection problem. We deployed three deep and six shallow learning algorithms for cyberbullying detection problems. The results show that bidirectional long-short-term memory is the most efficient method for cyberbullying detection, in terms of accuracy and recall.