Data used for TransBERT: A Three-stage pre-training Technology for Cyberbullying Detection

The rate at which cyberbullying is spreading across social media platforms is alarming. As much as using the internet opens us to the world it also comes with its risks like cyberbullying. Cyberbullying can be done by victimizing the victims through sending proactive messages or sensitive images or text. Detection of such messages texts or images can be very difficult on such large social media platforms and may sometimes lead to false detection. Although there exist various ways to detect cyberbullying, it keeps on rising. Social platforms such as Twitter, Facebook, Instagram, and Snapchat have created a conducive environment for cyberbullying to grow. A three-stage pre-training model, TransBERT (Transferable Bidirectional Encoder Representations Transformers), has achieved better results when it comes to the end of story prediction. TransBERT as a subcategory of BERT (Bidirectional Encoder Representations Transformers) (Li et al., 2019) has also performed well when it came to language understanding tasks and also cyberbullying detection (Paul & Saha, 2020; Yadav et al., 2020b). In this research, we present an application of the TransBERT model for cyberbullying detection on a tweeter dataset that is available to the public.