Realtime Cyberbullying Detection on Telegram: A Machine Learning Approach with Cyberbot

Cyberbullying is a common problem in the digital world we live, that puts people at serious psychological and emotional risk, particularly kids and teenagers. Social media platforms, which are widely used for communication and engagement, have given rise to a fertile environment for instances of cyberbullying. Establishing a safer online environment requires both identifying and combating cyberbullying. In this paper, we developed and evaluated three models: a binary classification model with 90% accuracy, a multiclass classification model with 95% accuracy, and a BERT classification model with 99% accuracy. These models were tested on real-world social network datasets, demonstrating high efficiency and accuracy in detecting cyberbullying. The superior performance of the model highlights its potential to significantly enhance cyberbullying detection systems, contributing to safer online environments and supporting efforts to combat harmful online behavior globally.