Expeditious Cyber-Bullying Detection Method Utilizing Compact BERT Models

Abstract

Since many people now use social media to disseminate hate, many researchers have been concentrating on the issue of detecting cyberbullying. in the last ten years. In this work, transfer learning is used to address this issue. We adjust our tiny BERT models using data from hate speech. We use the Focal Loss function to address the data's class imbalance. With this method, we were able to obtain cutting-edge outcomes on the hate speech dataset, including 0.91 precision, 0.92 recall, and 0.91 F1-score. Additionally, we demonstrate using our transfer learning pipeline that the more compact BERT models are much faster at detecting cyberbullying and are appropriate for real-time applications.