Liardetector: a linguistic-based approach for identifying fake news

Abstract

Due to the existing Web infrastructure and the popularity of social media platforms, it is easy to share information in large scale. Although this online scenario brings benefits to the society, it also favors malicious groups that propagate misinformation (e.g., alternative facts, fake news) on the Web, causing damages that range from affecting the reputation of public entities (companies, celebrities) to interfering on political process. In this work, we propose a novel classification approach based on linguistic patterns for identifying fake news. Our approach reduces the dimensionality of the feature space by encoding probability distributions of tokens (e.g., words) as Shannon entropy and Jensen-Shannon divergence values. We report experimental results using multiple data sets, which show that our approach is a win-win solution that improves efficacy and efficiency. Compared to the baseline, our approach uses four orders of magnitude less features, and achieve a gain up to 74.3% of F1-score.