Automation in Social Networking Comments With the Help of Robust fastText and CNN

Social networking and online conversation platforms provide us with the power to share our views and ideas. However, nowadays on social media platforms, many people are taking these platforms for granted, they see it as an opportunity to harass and target others leading to cyber-attack and cyber-bullying which lead to traumatic experiences and suicidal attempts in extreme cases. Manually identifying and classifying such comments is a very long, tiresome and unreliable process. To solve this challenge, we have developed a deep learning system which will identify such negative content on online discussion platforms and successfully classify them into proper labels. Our proposed model aims to apply the text-based Convolution Neural Network (CNN) with word embedding, using fastText word embedding technique. fastText has shown efficient and more accurate results compared to Word2Vec and GLOVE model. Our model aims to improve detecting different types of toxicity to improve the social media experience. Our model classifies such comments in six classes which are Toxic, Severe Toxic, Obscene, Threat, Insult and Identity-hate. Multi-Label Classification helps us to provide an automated solution for dealing with the toxic comments problem we are facing.