Exploring the Efficacy of Deep Learning Models for Multiclass Toxic Comment Classification in Social Media Using Natural Language Processing

Our thesis focuses on developing a deep learning-based toxic comment classifier. The classifier will be used to find and report potentially unpleasant or hazardous information on websites like social media, discussion boards, and comment sections. We want to encourage a welcoming and safe online community and stop the growth of online abuse such as cyberbullying and hate speech. Long Short-Term Memory (LSTM) and Hybrid LSTM-CNN (Convolutional Neural Network and LSTM based Approach) are two algorithms that the classifier uses to categorize the comments depending on their level of toxicity, such as threats, obscenity, insults, and identity-based hatred. The classifier’s input data came from Kaggle and underwent a number of pre-processing processes, including lemmatization and normalizing the text data.