Automatic Recognition of Cyberbullying in the Web of Things and social media using Deep Learning Framework

Abstract

The Web of Things (WoT) is a network that facilitates the formation and distribution of information its users make. Young people nowadays, digital natives, have no trouble relating to others or joining groups online since they have grown up in a world where new technology has pushed communications to a nearly real-time level. Shared private messages, rumours, and sexual comments are all examples of online harassment that have led to several recent cases worldwide. Therefore, academics have been more interested in finding ways to recognise bullying conduct on these platforms. The effects of cyberbullying, a terrible form of online misbehaviour, are distressing. It takes several documents, but the text is predominant on social networks. Intelligent systems are required for the automatic detection of such occurrences. Most previous research has used standard machine-learning techniques to tackle this issue. The increasing pervasiveness of cyberbullying in WoT and other social media platforms is a significant cause for worry that calls for robust responses to prevent further harm. This study offers a unique method of leveraging the deep learning (DL) model binary coyote optimization-based Convolutional Neural Network (BCNN) in social networks to identify and classify cyberbullying. An essential part of this method is the combination of DL-based abuse detection and feature subset selection. To efficiently detect and address cases of cyberbullying via social media, the proposed system incorporates many crucial steps, including preprocessing, feature selection, and classification. A binary coyote optimization (BCO)-based feature subset selection method is presented to enhance classification efficiency. To improve the accuracy of cyberbullying categorization, the BCO algorithm efficiently chooses a selection of key characteristics. Cyberbullying must be tracked and classified across all internet channels, and Convolutional Neural Network (CNN) is constructed. With a best-case accuracy of 99.5% on Formspring, 99.7% on Twitter, and 99.3% on Wikipedia, the suggested algorithm successfully identified the vast majority of cyberbullying content.