Fake News Detection Using Bayesian Inference

Given the huge volume of information available on social media, making a distinction between false information and a real one is a challenging task. In fact, several statistical models dealing with this problem are based on multinomial distributions. However, a new family of distributions that is an exponential family approximation to the Dirichlet Compound Multinomial (EDCM) has been introduced to be more adjustable to high-dimensional data and to overcome the drawbacks of the multinomial assumption. Thus, in this paper, we tackle the problem of fake news detection using finite mixture models of EDCM distributions. In particular, we develop a Bayesian approach based on Markov Chain Monte Carlo and Metropolis-Hastings algorithm for the learning of these mixture models. The proposed method is validated via extensive simulations and a comparison with multinomial-based mixture models is provided.