Effective Content Moderation: A Framework for Toxic Comment Analysis Using Machine Learning

This work showcases toxic comment analysis using Natural Language Processing to estimate toxicity levels of usergenerated comments on a scale from 0 to 1. By quantifying toxicity, the work addresses maintaining positive interactions in online environments. The benchmark output is the toxicity score as a regression task, with binary classification labeling comments as toxic (score >0.5) or non-toxic (score ≤0.5) for additional insights. This research leverages the datasets with associated toxicity levels, making it suitable for supervised machine learning. In this work, we combine the strengths of TF-IDF and Bag of Words(BoW) in proposing optimized SGD Regressor and Decision Tree based model for toxic comment classificattion.The model achieved an accuracy of 96%, along with AUC of 92%. It outperformed previously implemented techniques and models. We can conclude that it emerges as a powerful technique to evaluate toxicity on social media, leveraging it in diverse linguistic and cultural contexts.