Detecção e análise de metáforas em fake news

Abstract

Metaphor is a figure of speech largely present both in day-to-day conversations and formal texts. Furthermore, it is believed that metaphors affect the acquisition of new information by forming a new conceptual structure for the target, similar to that of the source characteristics it is being compared to. Given that fake news, a phenomenon that has gained academic and social attention, is used to confuse and manipulate readers, we seek to investigate whether metaphors are one of the methods used to bias the content of news articles. For that purpose, we perform the task of automatically detecting metaphors through a Long Short-Term Memory model and analyze the emotions associated with the resulting annotations for specific entities (i.e., Hillary Clinton and Donald Trump). We have observed that fake news present the highest concentration of metaphors by words in the body of its articles and that the emotions of the metaphor used to describe each candidate differ according to the type of news. We believe that studying the metaphors and the emotions they convey could be used in addition to the current methods being employed to detect fake news in order to improve the classification performance.