Cyberbullying Detection using Deep Learning Models in Bengali Language

Social media has developed into a very powerful tool for teamwork, communication, and idea exchange. Due to the anonymity offered by these platforms, verbal abuse and cyberbullying cases have, sadly, proliferated globally. Researchers and academics from all around the world have conducted studies to create methods to automatically detect cyberbullying in response to this disturbing epidemic. Unfortunately, the research that is now accessible primarily focuses on more advanced languages, leaving low-resource languages like Bengali with a substantial research gap. As a result, in this study, we present a method for detecting cyberbullying in Bengali using Deep Learning Models such as Bidirectional Long Short-Term Memory (Bi-LSTM) and Convolutional Neural Network (CNN). Furthermore, we utilised the Bengali Cyber Bullying public dataset from kaggle.com to illustrate this. Also, when we examined the outcomes of the Bi-LSTM and CNN models, we discovered that CNN performed better. The precision, recall, and F1 score produced by the CNN model are on average 0.827, 0.850, and 0.855.