Investigation of deep neural network architecture for cyberbullying detection over social media

Introduction: Nowadays, there has been a significant increase in cases of cyberbullying on digital devices and platforms such as Facebook, Instagram, Snapchat, and TikTok.Problem: Many state-of-the-art approaches have been introduced for the detection of cyberbullying activities. However, the affordability of high-quality data resources, along with restrictions on their access, limits the applicability of these state-of-the-art approaches.Objective: The detection of cyberbullying activities is of societal importance and has gained increasing prominence in research.Methodology: In this paper, we explored convolutional neural networks for cyberbullying detection (CNN-CBD) architecture for the classification task and reported their performance on real-world databases such as Twitter, Wikipedia, and Formspring. We also compared the CNN-CBD performance with baseline machine learning (ML) models. Various issues regarding the handling of real-world databases and the selection of the most suitable deep neural network (DNN) model are reported and discussed in detail.Results: Experiments showed that the proposed CNN-CBD model outperformed traditional ML algorithms in cyberbullying detection, achieving an accuracy of 97%.Conclusions: We concluded that the proposed CNN-CBD model outperformed the existing baseline models.