Fake News Detection using Deep Markov Random Fields

Proceedings of NAACL-HLT 2019, pages 1391–1400 Minneapolis, Minnesota, June 2 - June 7, 2019. c⃝2019 Association for Computational Linguistics 1391 Fake News Detection using Deep Markov Random Fields Duc Minh Nguyen⋆†, Tien Huu Do⋆†, Robert Calderbank‡, Nikos Deligiannis⋆†, ⋆Vrije Universiteit Brussel, Pleinlaan 2, B-1050 Brussels, Belgium †imec, Kapeldreef 75, B-3001 Leuven, Belgium ‡Duke University, Durham, North Carolina ⋆†{mdnguyen, thdo, ndeligia}@etrovub.be ‡robert.calderbank@duke.edu Abstract Deep-learning-based models have been suc- cessfully applied to the problem of detecting fake news on social media. While the corre- lations among news articles have been shown to be effective cues for online news analy- sis, existing deep-learning-based methods of- ten ignore this information and only consider each news article individually. To overcome this limitation, we develop a graph-theoretic method that inherits the power of deep learn- ing while at the same time utilizing the cor- relations among the articles. We formulate fake news detection as an inference problem in a Markov random field (MRF) which can be solved by the iterative mean-field algorithm. We then unfold the mean-field algorithm into hidden layers that are composed of common neural network operations. By integrating these hidden layers on top of a deep network, which produces the MRF potentials, we obtain our deep MRF model for fake news detection. Experimental results on well-known datasets show that the proposed model improves upon various state-of-the-art models. 1 Introduction The term “fake news” refers to news articles that is intentionally and verifiably false (Shu et al., 2017). The problem of fake news has existed since the ap- pearance of the printing press, but only gained a lot of momentum and visibility during the age of so- cial media. This is due to the large audience, easy access and fast dissemination mechanism of social media, where more and more users are consuming news in a daily basis (Shu et al., 2017). Tradi- tional methods for verifying the veracity of news that rely on human experts, despite being reliable, do not scale well to the massive volume of news nowadays. This renders the automatic detection of fake news on social media an important problem, w23 = 5 w14 = 3 w13 = 3 w34 = 3 a1 a2 a3 a4 Figure 1: Modeling the relationship among news arti- cles (or events): the dash lines represent engagements of users to articles, the solid lines represent the rela- tionships between articles with weights determined by the number of common engaged users. drawing a lot of attention from both the academic and industrial communities. The recent literature has witnessed the success of deep learning models in detecting fake news on social media (Ma et al., 2016; Yu et al., 2017; Ruchansky et al., 2017; Rashkin et al., 2017; Ma et al., 2018a; Kochkina et al., 2018). By lever- aging the capability of deep networks in learn- ing high-level representations, these models have achieved state-of-the-art performance on various benchmark datasets. Nevertheless, one limitation of existing deep-learning-based methods is that they often ignore the correlations among news ar- ticles, which …

Content truncated to 500 words