Fake News Dataset with In-content Annotations and Detailed Lying Excerpts

<p>This dataset contains 95 fake news collected from two Brazilian fact-checking services (E-farsas and Boatos). We carefully annotated some excerpts of the fake news to enable deeper analyses of their falsehood. From these annotations, we divided the news into three groups: real news, totally fake news and fake news but which contain only a few lying snippets.</p><p>The annotated fragments were grouped based on four categories of falsehood: untrue, unverifiable fact, incorrectly named entity, and exaggeration. Other parts of the fake news, such as the passages around the fragments, were separated to add more information to the results of this research.</p>