Abstract

Social media platforms have seen a major boom in the past decade. Apart from entertainment and establishing social connections, the use of social media to read news articles has become prevalent. News has not only become less costly but also fast and accessible. However, accessing news via social media has its own demerits. It is next to impossible for the end users to establish the validity of the news articles. Manual review of each and every article on the internet is neither quick nor economically feasible, considering the vast amount of news articles that are published on day-to-day basis. Hence, there arises a need for quick and highly accurate machine learning models to detect the fake news articles. In this paper, we propose a method to detect fake news via Graph based Markov chains. We initiate the classification process by first segregating the fake news articles from the real ones, then training two separate Markov Chains for both the classes, and finally calculating the probability that a news article was generated by a given Markov Chain. Our approach of establishing similarity between a text sequence and a Markov chain relies on probabilistic and statistical calculations, attained by performing random walks on the Markov chains. Such an approach has proven to provide a high accuracy on multiple datasets.