HATE SPEECH DETECTION USING MULTI CHANNEL CONVOLUTIONAL NEURAL NETWORK

Abstract

Web one of the modes of availability that is accessible at the doorstep, with admittance to the web one gains admittance to many online stages. An increment in the utilization of these stages gives us a few advantages just as certain downsides. One of such disadvantages is hate speech. Hate speech is a subject of worry for online media stages. With powerfully expanding datasets manual mediation of posts is very inconceivable or will be tedious. Hate speech detection should be an automated task to distinguish hate speech from the provided input. In this paper, we have implemented a deep learning model multi-channel convolutional neural network (MCCNN). The model consists of 3 channels of Convolutional Neural Network. Each channel is merged and connected to a fully connected layer from where the final output is obtained. We have compared our model with a single-channel convolution neural network and results have shown that MCCNN outperformed simple CNN. The accuracy and F1-score achieved by our model are 95.49 and 93.93 for dataset D1 and for dataset D2 97.85% and 95.74% respectively.