The benchmarks section lists all benchmarks using a given dataset or any of
its variants. We use variants to distinguish between results evaluated on
slightly different versions of the same dataset. For example, ImageNet 32⨉32
and ImageNet 64⨉64 are variants of the ImageNet dataset.
The Horne 2017 Fake News Data contains two independed news datasets:
Buzzfeed Political News Data:
News originally analyzed by Craig Silverman of Buzzfeed News in article entitled " This Analysis Shows How Viral Fake Election News Stories Outperformed Real News On Facebook."
BuzzFeed News used keyword search on the content analysis tool BuzzSumo to find news stories
Post the analysis of Buzzfeed News, the authors collect the body text and body title of all articles and use the ground truth as set by Buzzfeed as actual ground truth.
This data set has fewer clear restrictions on the ground truth, including opinion-based real stories and satire-based fake stories. In our study, the authors manually filter this data set down to contain only "hard" news stories and malicious fake news stories. This repository contains the whole dataset with no filtering.
Random Political News Data:
Randomly collected from three types of sources during 2016.
Sources ground truth determined through: Business Insider’s “Most Trusted” list and Zimdars 2016 Fake news list
Sources:
Real: Wall Street Journal, The Economist, BBC, NPR, ABC, CBS, USA Today, The Guardian, NBC, The Washington Post
Satire: The Onion, Huffington Post Satire, Borowitz Report, The Beaverton, Satire Wire, and Faking News
Fake: Ending The Fed, True Pundit, abcnews.com.co, DC Gazette, Liberty Writers News, Before its News, InfoWars, Real News Right Now