Towards fake news detection in Portuguese : New dataset and a claim-based approach for automated detection

Abstract

The spread of non-veridic information is a longstanding problem that has affected society ever since the advent of communication. The emergence of the Internet and Social Media have aggravated this issue, leading to a higher degree of influence of misinformation in peoples opinions and lives and therefore a higher impact on contemporary events, such as the 2016 US Elections. This led to the coining of the term Fake News, which has been widely used describe this recent phenomenon, and prompted it’s study by many fields of knowledge, and in the context of many languages. In the field of Computer Science, the main concern is the outstanding problem of automated fake news detection, which has barely been explored in the context of lusophone countries. One of the reasons for this, is the lack of content - datasets - which are required for work to be done on the subject. This work aims to provide a new labeled dataset for the problem of fake news detection in Portuguese, with news claims gathered from unbiased and non-partisan fact-checking sources, and apply methods of text-classification which have been proved to work on fake news, in order to validate if news can be classified as fake based solely on their claim. Apart from existing methods, this work also attempts a novel classification method, using the named entities extracted from the claim as a feature for classification.