Localization of Fake News Detection via Multitask Transfer Learning

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 2596–2604 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 2596 Localization of Fake News Detection via Multitask Transfer Learning Jan Christian Blaise Cruz, Julianne Agatha Tan, and Charibeth Cheng Center for Language Technologies (CeLT), De La Salle University, Manila 2401 Taft Ave, Malate, Manila, Philippines {jan christian cruz, julianne tan, charibeth.cheng}@dlsu.edu.ph Abstract The use of the internet as a fast medium of spreading fake news reinforces the need for computational tools that combat it. Techniques that train fake news classifiers exist, but they all assume an abundance of resources including large labeled datasets and expert-curated corpora, which low-resource languages may not have. In this work, we make two main contributions: First, we alleviate resource scarcity by constructing the first expertly-curated benchmark dataset for fake news detection in Filipino, which we call “Fake News Filipino.” Second, we benchmark Transfer Learning (TL) techniques and show that they can be used to train robust fake news classifiers from little data, achieving 91% accuracy on our fake news dataset, reducing the error by 14% compared to established few-shot baselines. Furthermore, lifting ideas from multitask learning, we show that augmenting transformer-based transfer techniques with auxiliary language modeling losses improves their performance by adapting to writing style. Using this, we improve TL performance by 4-6%, achieving an accuracy of 96% on our best model. Lastly, we show that our method generalizes well to different types of news articles, including political news, entertainment news, and opinion articles. Keywords: Statistical and Machine Learning Methods, Language Modelling, Document Classification and Text Categorization 1. Introduction There is a growing interest in research revolving around au- tomated fake news detection and fact checking as its need increases due to the dangerous speed fake news spreads on social media (P´erez-Rosas et al., 2018). With as much as 68% of adults in the United States regularly consuming news on social media1, being able to distinguish fake from non-fake is a pressing need. Numerous recent studies have tackled fake news detection with various techniques. The work of Bourgonje et al. (2017) identifies and verifies the stance of a headline with respect to its content as a first step in identifying potential fake news, achieving an accuracy of 89.59% on a publicly available article stance dataset. The work of Karimi et al. (2018) uses a deep learning approach and integrates mul- tiple sources to assign a degree of “fakeness” to an arti- cle, beating representative baselines on a publicly-available fake news dataset. More recent approaches also incorporate newer, novel methods to aid in detection. The work of Conforti et al. (2018) handles fake news detection as a specific case of cross-level stance detection. In addition, their work also uses the presence of an “inverted pyramid” structure as an indicator of real news, using a neural network to encode a given article’s structure. While these approaches are valid and robust, most, if not all, modern fake news detection …

Content truncated to 500 words