Comparative analysis for toxic comment classification

Abstract

Due to the development in e-commerce, social media channels like twitter and Facebook flood of Information is poured into the Internet, while this is going to affect the quality of the people since these texts may include many levels of toxicity which results in online harassment, bullying, etc., There is paramount importance to monitor this data. Research and Industrial communities are trying to build an effective model to classify these toxic comments. Still, these attempts are not reaching sufficient levels. Luckily In recent days, there are advancements in hardware and data management techniques like Big Data that encourages computational approaches like Deep learning approaches which can improve the text classification. This paper focuses primarily on, various approaches for short toxic comment classification and the comparative experiments are set to evaluate their accuracy based on Area Under Curve (AUC) metric from “Wikipedia Talk Page Comments annotated with toxicity reasons” Dataset. Based on our experimental analysis, bidirectional GRU performed better as compared to other existing classification algorithms. These findings will be used as the benchmarking results for improvement of Toxic Comment Classification.