A small manually fact-checked fake news dataset

A balanced and cleaned fake news dataset with fact-checked labels suitable as a test set for fake news classification tasks.

Dataset information

• Consists of in total 434 articles, 217 of each class
• Articles are mainly from 2017 and 2018, but some are from before 2017
• The articles in this dataset have been manually cleaned to remove empty articles/titles, advertisement etc.

The dataset is assembled from three different datasets:

1. FakeNewsNet (The Politifact dataset) - 217 fake and 73 real articles
2. MisInfoText (The Snopes dataset) - 58 real articles
3. Manually gathered articles from Snopes.com - 86 real articles

The dataset contains the following features:

• id - original label from the origin dataset
• label - 0 = real, 1 = fake
• url - url of the article
• title - article title
• content - article content

Disclaimer

This dataset was assembled as part of a master thesis, and we can not vouch for the correctness of all labels.