Novel Pipeline to Detect and Correct Cyberbullying
by Nehal Singh
Category: Computer Science
Abstract – In this work, a novel pipeline to detect and correct cyberbullying is proposed. The cyberbullying detection model utilizes sentimental, semantic, and syntactic features extracted through topic modelling, an emotion detection random forest model, and the term frequency. These features are inputted into an artificial neural network that can predict cyberbullying at an accuracy rate of 92%.. The text correction model implements an encoder-decoder architecture with a single-head cross-attention block, bi-directional recurrent neural network, and recurrent neural network. This model returned accuracies of approximately 100% and converted inappropriate text such as “racism is ok” to “don’t let ppl who bully you bring you down!”.