Detecting Toxic Comments Using Convolutional Neural Network Approach

Abstract

In the most significant issue now plaguing social networking platforms and online communities is toxicity identification. Therefore, it is necessary to create an automatic hazardous identification system to block and restrict individual from certain online environments. We introduce multichannel Convolutional Neural Network (CNN) approach in this paper for the detection of toxic comments in a multi-label context. With the help of pre-trained word embeddings, the suggested model produces word vectors. Also, to model input words with long-term dependency, this hybrid model extracts local characteristics using a variety of filters and kernel sizes. Then, to forecast multi-label categories, we integrate numerous channels with three layers as fully linked, normalization, and an output layer. The results of the experiments show that the suggested model performs where we are presenting the fresh modeling CNN approach to detect the toxicity of textual content present on the social media platforms and we categorized the toxicity into positive and negative impact on our society.