Comparative study of toxic comments classification using machine learning algorithms

Abstract

The rapid growth of information technology and the disruptive transformation of social media have happened in recent years. Websites like Facebook, Twitter, Instagram, where people can express their thoughts or feelings by posting text, photos or videos, have become incredibly popular. But unfortunately, it has also become a place for hateful activity, abusive words, cyberbullying and anonymous threats. There are many existing works in this field but those are not fully successful yet to provide accuracy in satisfactory level. In this work, we employ natural language processing (NLP) with convolution neural networking (CNN), extreme gradient boosting (XGBoost) and support vector machine (SVM) for segmenting toxic comments at first and then classifying them in six types from a large pool of documents provided by Kaggle’s regarding Wikipedia’s talk page edits. Using this dataset, the hamming score of CNN model is 89% ,XGBoost model is 87% and SVM model is 84%.