Detecção de viés ideológico de portais de notícias na Web

Abstract

Nowadays, websites or news portals are the main sources of information to most people. However, like traditional media, these vehicles can have a bias in the way they report news, favoring an ideology of interest. Combined with social media and the ease of spreading this type of content, this fact strongly contributes to polarization, hate crimes and other consequences in public opinion. To make the information more transparent to the public, it is necessary to develop methods to characterize the ideological orientation/leaning of these portals automatically. Recent approaches are not exactly suitable for this problem, as they mostly depend on external sources, generating inaccurate results otherwise. Therefore, in this work, we present methods to detect ideological/political bias in news portals based only on news articles from these portals, without any external sources. We developed two approaches: exploring hyperlinks and textual content. The objective is to demonstrate the efficiency and effectiveness of this strategy compared to the current literature. As a result, we show that an approach based on hyperlinks is capable of detecting ideological biases in a polarized scenario through a method based on citation patterns. In addition, we present an approach based on textual content associated with Information Theory concepts and show that the method is able to overcome a more traditional baseline, obtaining almost twice the accuracy/F1 in three datasets and three distinct classification tasks (bi-class and multi-class), while employing a set of only four features (against 282 employed by the baseline) when detecting different levels of ideological bias in news portals.