Detection of insulting comments in online discussion

Abstract

With time, people have graduated towards expressing themselves through the medium of the Internet, specifically over social media platforms. It creates scenarios where a user may, knowingly or unknowingly, make a reference or comment which may be derogatory to an individual and/or a section of society. It may hinder observant from participating in the conversation or even stop visiting the website altogether, thereby hurting the prospects of the website owner. A human may easily detect such infringements; however it is a huge pursuit for a computer. In this paper, we present a text classification method to classify the comments as insulting or otherwise. For this purpose, we extract features using various methods and enrich them using k-skip-n-grams to achieve a good set of features for the task. Further, feature selection is applied to obtain a subset of relevant features. Finally, a competitive and collaborative analysis of five different machine learning methods (classifiers) is presented to show that a collaborative model is a clear winner. It is a step towards making a machine learning based automated system to detect the insulting comments in the conversations.