Measuring the Impact of Readability Features in Fake News Detection

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 1404–1413 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 1404 Measuring the Impact of Readability Features in Fake News Detection Roney L. S. Santos1, Gabriela Wick-Pedro2, Sidney Leal1, Oto A. Vale2, Thiago A. S. Pardo1, Kalina Bontcheva3, Carolina Scarton3 1 Interinstitutional Center for Computational Linguistics (NILC), University of S˜ao Paulo, S˜ao Carlos, Brazil 2 Department of Languages, Federal University of S˜ao Carlos, S˜ao Carlos, Brazil 3 Department of Computer Science, University of Sheffield, Sheffield, United Kingdom roneysantos@usp.br, gwpedro@estudante.ufscar.br, sidleal@gmail.com, otovale@ufscar.br, taspardo@icmc.usp.br, {k.bontcheva, c.scarton}@sheffield.ac.uk Abstract The proliferation of fake news is a current issue that influences a number of important areas of society, such as politics, economy and health. In the Natural Language Processing area, recent initiatives tried to detect fake news in different ways, ranging from language-based approaches to content-based verification. In such approaches, the choice of the features for the classification of fake and true news is one of the most important parts of the process. This paper presents a study on the impact of readability features to detect fake news for the Brazilian Portuguese language. The results show that such features are relevant to the task (achieving, alone, up to 92% classification accuracy) and may improve previous classification results. Keywords: Fake news, Readability Assessment 1. Introduction The term “Fake News” can be defined as fabricated infor- mation that mimics news media content in form but not in organizational process or intent (Lazer et al., 2018), as news articles that are intentionally and verifiably false, and could possibly mislead readers (Allcott and Gentzkow, 2017), and simply as low quality news with intentionally false infor- mation. The task of detecting fake news is defined as the prediction of the chances of a particular news article being deceptive (Rubin et al., 2015). According to Rubin et al. (2015), two main categories of fake news detection methods currently stand out: 1) net- work approaches, in which network information, such as message metadata or structured knowledge network queries can be harnessed to provide aggregate deception measures; and 2) linguistic approaches, in which the content of de- ceptive messages is extracted and analyzed to associate language patterns with deception. In network approaches, there are fact-checking (or content verification) models, in which the use of other techniques such as graph theory, complex networks and question answering, being able to link the subjects of an statement that refers to various fac- tual characteristics, such as “is a”, “member of” and “is married to”, among others (Ciampaglia et al., 2015). In the case of linguistic approaches, it is hypothesized that fake news show linguistic clues that make them detectable when compared to true news, since it is believed that the deceiver unconsciously reflects its deceptive behavior and readabil- ity issues in the text. The choice of language clues (i.e., features) that can be used to classify news as true or false vary considerably. Previous work uses n-grams, lexical measures …

Content truncated to 500 words