Cyberbullying Dataset Collection for an Enhanced Automatic Detection on Twitter Using Deep Learning Techniques

Cyberbullying has become a serious issue in the digital world. As cyberbullying occurs in an online environment it can reach victims at any location or time unlike traditional bullying that occurs face-to-face. In addition, cyberbullying can have severe effects on the mental health and physical wellbeing of victims, leading to harmful actions against themselves such as anxiety, depression, and suicidal thoughts. Due to the vast volume of content generation on social media platforms as well as the complexity of the language, it presents a significant challenge to manually detecting instances of cyberbullying. Given the scalability issue in this manual detection, hence the urgent need for the development of automated detection systems. In this study, we aim to enhance detection of cyberbullying by first generating a recent public dataset collected from Twitter. Then, we trained a Bi-Directional Long Short-Term Memory (Bi-LSTM) to automatically detect cyberbullying. Experimental results demonstrate that the proposed Bi-LSTM model surpasses traditional methods, contributing significantly towards the development of robust solutions for mitigating online harassment.