An ideal deeply contextual prediction model for an effective prediction and mitigation of cyberbullying in social networks

Social media have evoked a host of problems, especially Twitter and Instagram, considering misconceived notions of free speech. Among the concerns that have been seen as crucial is the impact of cyberbullying; it is a global issue affecting an individual and society in general. Though various methods have been suggested for deterring cyberbullying, most of these depend on victim interaction and are hence ineffective in the task of independently identifying and preventing cyberbullying. Within this context, the research introduces a new, innovative framework that is ideal and deeply contextual in terms of prediction, thus creating an approach to detecting cyberbullying on social networks without directly engaging with any victim. Consequently, the key contribution of this work is developing the Embedded Residual Recurrent Convolution Network, a hybrid deep learning architecture for the detection of cyberbullying data with more precision and accuracy. Further, this is optimized by a Tenacious Projectile Optimizer for minimizing computational time and improving classifier performance. In this regard, the novelty of the proposed I-DCP framework is established in its superior performances on different social media datasets, with an average accuracy of over 99% and an execution time as low as 13 s, hence highly efficient for real-time detection of cyberbullying. These represent important advances compared to previous methodologies and constitute a strong and scalable response in the fight against cyberbullying.