A Gaussian Naive Bayesian Classifier for Fake News Detection in Bengali

Abstract

With the advent of modern digital technology and reach of digitized contents manipulation of facts turned to fake news and their impact is widespread than ever. The intent is often to manipulate consents on religious, political, financial and other serious matters within the social and state context and create a nuisance and spread violence even wage wars. However, common people are not able to distinguish between fake and real news. Often the dubious nature of the fake news makes us even to suspect the real news. With the progress made in natural language processing it has become interesting to seek for knowledge or patterns in the generation of fake news and thus find better predictive ways to fake news to differentiate it from real news. In this paper, we propose a machine learning based fake news detection method in Bengali. Our proposed method uses a novel dataset created for the purpose and a Gaussian Naive Bayes Algorithm. The algorithm uses TF-IDF based text features and Extra Tree Classifier for feature selection. In addition to this, we have performed comprehensive analysis on different machine learning algorithms and on features.