Misinformation, radicalization and hate through the lens of users

• Autor(a) principal: Manoel Horta Ribeiro
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Institucional da UFMG
• Texto Completo: http://hdl.handle.net/1843/46708
• topic: Social networks; Hate speech; Misinformation; Radicalization; Computação - Teses; Redes sociais on-line - Teses; Desinformação - Teses; Fake news - Teses

The popularization of Online Social Networks has changed the dynamics of content creation and consumption. Barriers to disseminate texts, images and videos became significantly lower than in earlier times, empowering users to create, with little resources, content with far-reaching impact. In this setting, society has witnessed an amplification in phenomena such as misinformation and hate speech. Recent research attempts to address these issues by studying hateful or fake content. Yet, robustly addressing these phenomena in this fashion is often not feasible. Consider for example the task of detecting hate speech. There can be two exact texts that, in different contexts (e.g. a rap lyric and a politician's speech), may be considered to be hateful or not. Moreover, there may even be that, given a text within a context, two individuals disagree on whether the content is hateful or not. This dissertation attempts to address these issues by taking another perspective: that of the user. Through the lens of users, it is possible to approach possibly fake or hateful content with surrounding context: political orientation, activity patterns, connections, etc. Furthermore, we can study more complex phenomenon, such as user radicalization, where one must study the trajectories of individuals -- or, more realistically, their online traces. In three case studies on social networks, we: (i) provide insight on how the perception of what is misinformation is altered by political opinion; (ii) propose a methodology to study hate speech on a user-level, showing that the network structure of users can improve the detection of the phenomenon; (iii) characterize user radicalization in far-right channels on YouTube through time, showing a growing migration towards the consumption of extreme content in the platform. Each case study contributes to their more specific subject: misinformation, hate speech and user radicalization. Yet, altogether, they advance a central argument: that studying users rather than the content itself is more productive to better understand (and eventually mitigate) ill-defined social phenomena such as hate speech and fake news.