You're allowed to use the following sources of data:

- https://github.com/several27/FakeNewsCorpus
- Any data you mined during the first phase of the project 

Since these two datasets (and the data you mined) have different label sets, it will be up to you to decide how to best bin them into 'True' and 'False' classes. The final (hidden) test set will only have 'REAL' and 'FAKE' labels. 

Below you can download the test set (from an unknown source), for which the labels have been removed. You're then supposed to submit predictions for this test set in the form of: ID,PREDICTION.

The predictions file that you submit for the test set should have the following header: "id,label". 

The liar dataset ( https://www.cs.ucsb.edu/~william/data/liar_dataset.zip) will be used for cross-domain testing.