Publication

Abstract :

In recent days, the online conversations have become a vibrant platform in expressing one’s opinion about the issues prevailing in the society. Increase of threats, abuses and harassment in social websites has stopped many from expressing themselves, and they give upon seeking different opinions due to the fear of being offended and cornered. This can be advocated with the advancements of machine learning and artificial intelligence, which offers a way to possibly classify the comments based on its level of toxicity and its nature. In this following work, we propose an application that classifies the comments based on the nature of toxicity. Here, the learning model is trained with datasets containing toxic comments, and when such comments are used, the prototype identifies and highlights it. This is carried out with the most popular data classifier called logistic regression in which the prototype and feature extraction is done using sklearn’s TfidfVectorizer module. Once the toxic comments are identified by the trained application, warning flag will be raised to the user and analogous assertive statements are provided as suggestion by the recommendation engine, in which item-to-item-based collaborative filtering algorithm is used. The obtained results justifies that logistic regression affords adequate evidence in identifying the toxic comments.