Wikipedia hoaxes

The master webpage for this project is hosted at the University of Maryland:
http://cs.umd.edu/~srijan/hoax/

Dataset information

Wikipedia has over 35,000,000 articles in over 290 languages. However, not all the articles are genuine. Hoax articles are purely fabricated articles that were created to mislead people.

In the paper cited below we study all actual and wrongly suspected hoaxes ever identified in the English version of Wikipedia. Most of them have been permanently deleted from Wikipedia's version history, so we had access to them only under a non-disclosure agreement. Therefore we are unable to publish the full dataset we work with in the paper. Instead, we publish a smaller dataset of hoaxes that are also publicly available (on websites such as Speedy Deletion Wiki or Deletionpedia), alongside an equally-sized set of non-hoaxes.

This dataset contains a set of 64 hoax articles that are publicly available, and have the following properties:

the article was patrolled,
the article was not deleted for at least one month, and
the article has at least 100 page views per day.

It also contains a set of 64 non-hoax articles that have the above three properties as the hoax articles. In addition, these non-hoax articles are selected such that (i) for each hoax, there is a non-hoax article that was created on the same day, and (ii) the two sets have similar appearance features (see Section 6 of the paper).

Sources (citations)

Srijan Kumar, Robert West, and Jure Leskovec: Disinformation on the Web: Impact, Characteristics, and Detection of Wikipedia Hoaxes. Proceedings of the International Conference on the World Wide Web, 2016.

Files

The dataset contains four folders:

hoax_markup_cleaned: This folder has 64 files, one for each hoax article. Each file is the raw content of the article, along with all its markup (template, hyperlinks, image, table)
nonhoax_markup_cleaned: This is the corresponding set for the non-hoax articles.
hoax_pretty_parsed: This folder has 64 files, one for each hoax article. Each file is processed from its version in the hoax_markup_cleaned file to make the article look like a Wikipedia article when viewed in a browser.
nonhoax_pretty_parsed: This is the corresponding set for the non-hoax articles.

File	Description	Size
wiki-hoaxes.zip	Content of hoax and non-hoax articles	1.0 MB