Media bias characterization in Brazilian presidential elections

Abstract

News media bias is commonly associated with framing information so as to influence readers judgments. One way to expose such bias is to compare different news outlets on the same stories and look for divergences. In this paper, we investigate news media bias in the context of Brazilian presidential elections by comparing four popular news outlets during three consecutive election years (2010, 2014, and 2018). We analyse the textual content of news stories in search for three kinds of bias: coverage, association, and subjective language. Coverage bias has to do with differences in mention rates of candidates and parties. Association bias occurs when, for example, one candidate is associated with a negative concept while another not. Subjective bias, in turn, has to do with wording that attempts to influence the readers by appealing to emotion, stereotypes, or persuasive language. We perform a thorough analysis on a large scale news data set where several of such biases are exposed.