A Hybrid Framework for Image Cyberbullying Recognition Using Transfer Deep Learning

Facebook, Twitter, and other social media sites have many positive uses, but they also have many negative aspects. Cyberbullying (CB) is a problem on various social media sites. Since each individual's reaction to cyberbullying is unique, the damage it does to victims' lives is difficult to put a price on. While some may view the message as natural, victims may perceive it as bullying. Finding the bully material in cyberbullying communications is quite difficult due of their ambiguity. There has been some study on using textual posts to solve this problem. Cyberbullying detection using images, on the other hand, has gotten comparatively less focus. The overarching goal of this study is to provide a framework for addressing the problem of cyberbullying on social media sites that makes use of images. Developing models is the initial use case for the convolutional-neural-network(CNN) that is based on deep-learning (DL). In the subsequent sections, this study employs transfer-learning(TL) models. The experimental results of varying the hyper-parameters verified that the model based on TL is the superior option for this issue. With a bestcase accuracy of 95.6%, the suggested algorithm successfully identifies the vast majority of cyberbullying posts. Combining DL with TL techniques yielded better results than applying either method alone, according to experiments conducted on an image dataset.