Cyberbullying Identification and Classification using Hybrid Techniques

Abstract

Now a days in digital society, cyberbullying is a vigorous and widespread issue affecting huge number of online users, mostly teenagers. However, increased usage of social media lead to increase in the rise of cyberbullying. Cyberbullying is an aggressive action usually carried out by some persons using any electronic items, against people who cannot protect themselves. The existing works distinguishes the bullies from the usual Instagram users by considering the user, network and text related attributes using classifiers. Most of the available cyberbullying detection methods are supervised but they have mainly two drawbacks such as labelling the data which often take more time and labour and Current guidelines for labelling may not useful for future instances because of evolving social networks and different language usage. In order to overcome those limitations, the proposed work is introduced for the unsupervised cyberbullying detection method. Our proposed detection method will extract linguistic attributes such as idioms, sarcasm, irony and active or passive voice. In addition, the representation learning network which learns the multimodal session representation and the multitask learning network which will simultaneously understand the bullying energy and then models the comments arriving times from Instagram dataset.