Rapid Cyber-bullying detection method using Compact BERT Models

摘要

Nowadays, many people use their social media platform to spread hate online and that is why the problem of cyber-bullying detection has been the focus of many researchers over the past decade. In this work, we tackle this problem with transfer learning. We use various compact BERT models and fine-tune them with hate-speech data. We incorporate Focal Loss function to handle class imbalance in the data. Using this approach, we were able to achieve state-of-the-art results of 0.91 precision, 0.92 recall and 0.91 F1-score on the hate-speech dataset. Additionally, using our transfer learning pipeline, we show that the more compact BERT models are significantly faster in detection and are suitable for real-time applications of cyber-bullying detection.