In today’s digital society, cyberbullying is serious and widespread issues affecting high number of Internet users, mostly teenager. However, increases in social media usage there is increase in the rise of cyberbullying. Cyberbullying is aggressive act carried out by some person using electronic forms of contact, repeatedly against people who cannot defend themselves. In the existing work distinguished the bullies from normal Instagram users by considering text, network and user related attributes using classifiers. Most existing cyberbullying detection methods are supervised and, thus, have mainly two key drawbacks such as labeling the data often take more time and labor and Current guidelines for labeling may not useful for future instances because of evolving social networks and different language usage. To address these limitations, this proposed work introduces for unsupervised cyberbullying detection method. The proposed detection method will be extract linguistic attributes such as idioms, sarcasm, irony and active or passive voice. In addition, a representation learning network that learns the multi-modal session representations and a multitask learning network will simultaneously estimate bullying energy and then models the comments arriving times from Instagram data set.

Cyberbullying Instagram Multimodal Hybrid Unsupervised