About Test Data attribute "atricle_url"

[huan liu]

Hi, 

In the public test set, I observed that in "article_url" there are more than 810 snopes, and they are all oocs. Is this also true in hidden test data?

[shivangi.tum]

We can't tell you what the hidden test looks like since that would be unfair.

But do ignore the article_url attribute to develop your algorithm, since it is only provided for reference to read the entire story about how/when/where/why the image was circulated and should not be used to make your decision. 

Also, a simple Ctrl+F for keyword snopes in the test file returns 485 matches, not sure how you're getting 810, please double check

Pro Tip: Your method would fail on hidden test set if you'd use article_url while developing your method, coz images can come from a lot of sources. 

[shivangi.tum]

Hi,

you were given the dataset for research and you're aren't allowed to share dataset snapshots here since this group is publicly accessible so I have to delete your message. 

On Wednesday, May 26, 2021 at 9:53:22 AM UTC+2 huan liu wrote:

Hi，

As you can see in the figure below, after Ctrl+F in the file test_data.json, I found 818 snopes