Detecção de viés ideológico em artigos de notícias utilizando Aprendizado Métrico Profundo e Representações Contextuais

Abstract

Given the rapid expansion of digital media and the widespread dissemination of information through online platforms, web news portals have become the primary source of information for a large portion of the population. The way this information is presented and framed can significantly influence public opinion, shaping the discourse around critical issues and ultimately impacting political decision-making processes. To enhance the transparency of informa- tion provided to the public, it is essential to develop solutions that enable the automatic characterization of ideological orientation in news articles. How- ever, recent approaches have proven only partially effective in this regard, as they often rely on external sources and multiple feature generation processes, which can lead to inaccuracies. In this context, this work proposes methods for classifying ideological bias in news articles based solely on the content of the articles from these portals. To achieve this, we explore a deep learning ap- proach that integrates different forms of textual representation, employing deep metric learning techniques. The objective is to demonstrate the effectiveness of this strategy compared to the existing literature. The proposed approach combines contextual and semantic learning with textual features, validating it on a dataset composed of over 30,000 news articles covering a wide range of topics