Toxic Comment Classification Using Bi-LSTM, CNN, and TinyBERT

Social Media is a vital aspect of daily life where individuals express their ideas and opinions, participate in social and political debates, share photos and make content on a wide range of subjects. However, social media platforms can also be a breeding ground for cyberbullying, hate speech, and harassment, which can lead to real-world harm. The toxic nature of many discussions on social media is one such negative side effect. This project aims to classify toxic comments using three machine learning models. The models used in this project are Bidirectional Long Short-Term Memory (Bi-LSTM), Convolutional Neural Networks (CNN), and Tiny Bidirectional Encoder Representations from Transformers (TinyBERT). The newly proposed model allows the user to choose the algorithm suitable for their respective platform. This model will deliver a multi-binary output where it is classified into the following labels: Toxic, Severely Toxic, obscene, Insult, Hate, and Threat. The metrics used to evaluate the performance of the models include recall, accuracy, and precision. Social media platforms can use the suggested method to automatically identify and filter toxic comments, enhancing the security and quality of online interactions.