“FRIENDLY CHAT”- A chat application with multi-headed classification models for identifying abusive levels and their comparative study

Abstract

Abstract An application that tracks the use of offensive and vulgar language on chats on social media networks in order to prevent users from becoming disrespectful may play a role in changing how users interact. "Friendly chat" is a play on a community messaging program, which allows users to address a range of topics that are totally unregulated. Once things start heating up a user may become abusive towards other users, the application can disperse the situation quickly by implementing protective measures to stop the abusive user in his tracks. Abuse is however not the only type of "toxic" material to which a user is exposed when communicating with other users. Hatred of identification, intimidation, and other forms of obscenity may also become a part of the discussion. To identify a post as toxic, we need to refer to a reference, using the "toxic comment" dataset consisting of a large number of Wikipedia comments which have been rated by human critics who classify the comment to fall under one or more of the above-mentioned categories. Integrate this model in the real time data that we receive from “Friendly Chat” to classify a user into one or more of these categories. Use techniques of classification and machine learning such as Naïve Bayes, LSTM and Binary relevance and chain classifiers models. Our application’s capable of detecting abusive users.