Is the News Deceptive? Fake News Detection using Topic Authenticity

Aviad Elyashar, Jorge Bendahan, Rami Puzis

In this paper, we propose an approach for the detection of fake news in online social media (OSM). The approach is based on the authenticity of online discussions published by fake news promoters and legitimate accounts. Authenticity is quantified using a machine learning (ML) classifier that distinguishes between fake news promoters and legitimate accounts. In addition, we introduce novel link prediction features that were shown to be useful for classification. A description of the processes used to divide the dataset into categories representing topics or online discussions and measuring the authenticity of online discussions is provided. We also discuss new data collection methods for OSM, describe the process used to retrieve accounts and their posts in order to train traditional ML classifiers, and present guidelines for manually labeling accounts. The proposed approach is demonstrated using a Twitter pro-ISIS fanboy dataset provided by Kaggle. Our results show that the method can determine a topic’s authenticity from fake news promoters, and legitimate accounts. Thus, the suggested approach is effective for discriminating between topics that were strongly promoted by fake news promoters and those that attracted authentic public interest.