Análise de viés na cobertura da imprensa baseada em conteúdo textual de notícias

Abstract

Media bias can be described as factors, political or ideological identification, internal to the media reflected in the news articles. One way to expose media bias is to compare news articles, from different news outlets, on the same event, and look out for differences. In this research we study four indicators of media bias: coverage, association, subjectivity and lan- guage bias. Language and association biases are propositions of this thesis, while coverage and subjectivity are existing biases in the reviewed literature, analyzed by new perspectives in this work. The coverage bias is related to the amount of space given by the news agency to an entity. The association bias occurs when, for example, an entity is associated with a nega- tive concept while another is not. The subjectivity bias, in turn, has to do with words that try to influence readers by appealing to emotions, stereotypes, or persuasive language. Finally, language bias is related to the inherent subjectivity of the language that we use to communi- cate. We apply the coverage, association and subjectivity bias indicators to a dataset of news about the Brazilian presidential elections, exposing biases from four popular news agen- cies in three consecutive elections (i.e., 2010, 2014 and 2018). In addition, we applied the subjectivity and language bias indicators in a multilingual context, exposing biases in news related to Venezuelan Crisis and the Syrian War written in Portuguese, German, English and Spanish.