Alghafiqe Alhloul and Abu Alam, Gloucestershire University, UK

The prevalence of cyberbullying toward minorities has been a global concern in the last decade. This concern reached a crescendo during the COVID-19 pandemic as many online users became more active on Twitter, using the social media to harass and threaten vulnerable groups. The after effects of COVID-19. Aside from the increase in technology use, there are other factors at play that are causing an increase in cyberbullying. For instance, when there is a major crisis like the one that COVID-19 brings, this puts everyone on edge, and kids are no exception. As a result, hostility toward others tends to increase along with self-preserving and self-defensive behaviours. In this work, we proposed a novel framework to detect cyberbullying on Twitter. This framework combined the attention layer and the convolutional pooling layer to extract cyberbullying-related keywords from users’ tweets efficiently. We probed the effectiveness of the proposed model using 47000 labelled tweets, which were categorized into cyberbullying classes such as age, ethnicity, gender, religion, type of cyberbullying, and non-cyberbullying. In this research we used two sets of combinations. In the first part we used the combination of CNN and ML models. In this structure we used convolutional layers as feature extractor and then we used ML models such as RF and LR , CNN-XGB, CNN-LSTM, CNN attention for classification.

Cyberbullying, Self-attention, Convolution network, Machine learning, Deep learning.