Re: Article 32
@DougS
I get what you're saying but that's not quite the point I was making. They identified the post as about sexual assault by having teams of low-paid, south-east asian workers read and assess the posts and classify them. This is how they found ones that were 'news-worthy' and I don't believe they are denying they do this.
Once that was done, they had a pile of messages that were judged by the worker bees to be about sexual assault - possibly specifically in the military, possibly only generally.
At any rate, the two competing claims are that they either used location data to identify those posts that came from military bases (as per the Guardian) or, according to Whisper, that it was done purely by looking at the text and assessing that. Given that these were already pre-selected by low-cost workers overseas (apparently), it's not unreasonable that, given the gravity of this particular subject, they had people vet them more carefully, examining every message.
If so, 'Article 32' could well fall into that category.
On the matter of automated keyword checking, it appears as though the app selects a background image from some database, based on words it finds. The photo on this particular post is an article about a US Marine operation. The only bit in the text that could have been used to select that is 'Article 32'. Perhaps 'promoted' but obviously that's not specifically military.
If the database used to select the images is, as you would expect, made up of pre-selected images (as they would have to be royalty/copyright-free) those images would have tags that can then be matched up with keywords in the text.
My point is that this text was, by the image, evidently AUTOMATICALLY picked as relating to the military. All you need to do is to search/cross-reference that information.
As I said, I disapprove of all of this but this post has been selected by the author to specifically contradict the statements of Whisper and I don't think it proves those statements false as it seems easy enough to reconcile what Heyward is claiming with this particular example.
To show him as wrong, you have to show a post that would not be able to be identifiable as from/about the military without solely by the text. I don't believe this example fits that criteria.