Analysis of the Subjectivity Level in Fake News Fragments

Abstract

The widespread of fake news is increasingly worrying society and demanding approaches for mitigation. Although many approaches have been proposed to fake news detection, there is still a lack of works that deeply investigate their structure. Our study has been motivated by two findings discussed in existing works: the first is the fact that fake news usually mix real and fake information to mislead readers; the second is that subjective language is a resource commonly exploited by fake news producers. Therefore, to better understand how fake news is structured, we perform an analysis on the way the subjective language is exploited in different situations inside the fake news documents. For this, we built a dataset by manually identifying fake parts of news articles, and we also propose tags to categorize the fragments in documents. The proposed tags categorize the fake news fragments according to their kind of falsehood in document. To reveal subjectivity nuances within the fragments, we use the Word Movers Distance and a set of subjectivity lexicons in the Portuguese language. Our results indicate that the fragmentation of news allows the identification of subjectivity markers that cannot be identified when considering the entire documents.