Evaluating Deep Neural Networks for Automatic Fake News Detection in Political Domain

Abstract

Fake news has become a hot trending topic after the latest U.S. presidential elections when Donald Trump took office. The political speech during the presidential campaign was plagued with half-truths, falsehoods, and click-baits, creating confusion for the voters. Several algorithms have been designed to tackle the automatic fake news detection problem, but some issues still remain uncovered. Some approaches address the problem from a perspective where the website reputation is used as part of their analysis. Typical algorithms take into account text patterns and statistics for automatic fake news detection. Commonly, the fake news detection problem is treated as a multi-class text classifier. This paper proposes several deep neural architectures to classify fake news in the political domain. Furthermore, we demonstrate that combining statements and credibility patterns of politicians are very important for detecting fake news in a deep neural network classifier. We have found that the information about the politician is very useful for any of the tested architectures.