Toxic Chat Synthesis, Reporting and Removing in Social Media using Hybrid Deep Learning Model

Following the COVID-19 epidemic, social media usage has increased dramatically among a wide range of users, from young people to the elderly. These platforms are essential venues for people to share ideas, demonstrate their abilities, and voice their opinions. But in addition to these advantages, there has been a noticeable rise in the spread of offensive and vulgar information. This is especially concerning because there are groups who deliberately make insulting remarks that target women, children, and the elderly in an effort to stifle peaceful speech. Even though there are reporting procedures in place, the lengthy process of filing complaints deters many users from doing so. Our research provides an automatic method made to identify, report, and delete offensive communications from the Telegram social media platform in order to address this problem. Some previous work to classify comments is already present. But it is only on English language. In this work, we are extending this to Malayalam, Manglish (Malayalam language written with English letters), Mixed (mix of Malayalam and Manglish) and English Languages. Apart from the above two, we have created a Hybrid Deep Learning model combining CNN and RNN and compared its performance with already existing models like Logistic Regression, SVM, KNN and BERT. Our project aims to create an open and courteous online community where meaningful conversations can flourish without the constant influence of offensive language by tackling toxic communication at its source. This automatic approach is a big step in the direction of making Telegram a happier and more encouraging place for users to interact online.