Automatic Detection of Fake News

Proceedings of the 27th International Conference on Computational Linguistics, pages 3391–3401 Santa Fe, New Mexico, USA, August 20-26, 2018. 3391 Automatic Detection of Fake News Ver´onica P´erez-Rosas1, Bennett Kleinberg2, Alexandra Lefevre1 Rada Mihalcea1 1Computer Science and Engineering, University of Michigan 2Department of Psychology, University of Amsterdam vrncapr@umich.edu,b.a.r.kleinberg@uva.nl,mihalcea@umich.edu Abstract The proliferation of misleading information in everyday access media outlets such as social me- dia feeds, news blogs, and online newspapers have made it challenging to identify trustworthy news sources, thus increasing the need for computational tools able to provide insights into the reliability of online content. In this paper, we focus on the automatic identification of fake con- tent in online news. Our contribution is twofold. First, we introduce two novel datasets for the task of fake news detection, covering seven different news domains. We describe the collection, annotation, and validation process in detail and present several exploratory analyses on the iden- tification of linguistic differences in fake and legitimate news content. Second, we conduct a set of learning experiments to build accurate fake news detectors, and show that we can achieve accuracies of up to 76%. In addition, we provide comparative analyses of the automatic and manual identification of fake news. 1 Introduction Fake news detection has recently attracted a growing interest from the general public and researchers as the circulation of misinformation online increases, particularly in media outlets such as social media feeds, news blogs, and online newspapers. A recent report by the Jumpshot Tech Blog showed that Facebook referrals accounted for 50% of the total traffic to fake news sites and 20% total traffic to reputable websites.1 Since as many as 62% of U.S. adults consume news on social media (Jeffrey and Elisa, 2016), being able to identify fake content in online sources is a pressing need. Until now, computational approaches for fake news detection have relied on satirical news sources such as “The Onion” (Rubin et al., 2016), viral news tracking websites such as BuzzFeed (Potthast et al., 2017) and fact-checking websites such as ”politiFact” (Wang, 2017) and “Snopes” (Popat et al., 2016). However, the use of these sources poses several challenges and potential drawbacks. For instance, using satirical content as a source for fake content can bring underlying confounding factors into the analysis, such as humor and absurdity. This is particularly the case for satirical news from “The Onion”, which has been used in the past to explore other text properties such as humor (Mihalcea and Strapparava, 2005) and irony (Wallace, 2015). Moreover, fact-checking websites are usually constrained to a particular domain of interest, such as politics, and require human expertise to verify the news claims making it difficult to obtain datasets that provide some degree of generalization over other domains (Chen et al., 2015). In this paper, we develop computational resources and models for the task of fake news detection. We introduce two novel datasets covering seven different domains. One of the datasets is collected by combining manual and crowdsourced annotation approaches, while the second …

Content truncated to 500 words