Towards automatic fake news detection in digital platforms: properties, limitations, and applications

• Autor(a) principal: Julio Cesar Soares dos Reis
• Data de Publicação: 2020
• Tipo de documento: Tese
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/34447; https://orcid.org/0000-0003-0563-0434
• topic: Computação – Teses; Mídia social; Fake news; Desinformação; Aprendizagem de máquina; Computação – Teses.; Mídia social - Teses.; Fake news - Teses.; Desinformação - Teses.; Aprendizagem de máquina - Teses

Digital platforms have dramatically changed the way news is produced, disseminated, and consumed in our society. A key problem today is that digital platforms have become a place for campaigns of misinformation that affect the credibility of the entire news ecosystem. The emergence of fake news in these environments has quickly evolved into a worldwide phenomenon, where the lack of scalable fact-checking strategies is especially worrisome. Thus, automatic solutions for fake news detection could be used as an auxiliary tool for fact-checkers to identify content that is more likely to be fake, or content that is worth checking. In this context, this thesis aims at investigating practical approaches for the automatic detection of fake news in digital platforms. First, we survey a large number of recent and related works as an effort to implement all potential features to detect fake news. We propose novel features and explore labeled datasets proposing new ones to assess the prediction performance of current supervised machine learning approaches. Our results reveal that these proposed computational models have a useful discriminative capacity for detecting fake news disseminated in digital platforms. We then propose an unbiased framework for quantifying the informativeness of features for fake news detection. As part of our proposed framework, we present an explanation of factors contributing to model decisions, thus promoting civic reasoning by complementing our ability to evaluate digital content and reach warranted conclusions. We also analyze features and models that can be useful for detecting fake news from different scenarios: the US and Brazilian elections. Finally, we propose and implement into a real system a new mechanism that accounts for the potential occurrence of fake news within data, significantly reducing the number of content pieces journalists and fact-checkers have to go through before finding a fake story.