3 ms·
> After reading all of the 1,000 replies by the user, the human then made the assignment based on the following criteria: (1) “Did the user’s replies entirely,
by Eisenstein 3y ago
> After reading all of the 1,000 replies by the user, the human then made
the assignment based on the following criteria: (1) “Did the user’s replies entirely, or almost entirely support one candidate?”; (2) “Did the user’s posts generally contain claims to support their arguments?”; and (3) “Did the user explicitly mention a tie to any campaign?”
If they can't see why these criteria are nonsensical then their entire premise should be discarded.
- lcnPylGDnU4H9OF 3y ago> why these criteria are nonsensical I don't conduct surveys or other research so it's not immediately obvious to me. I would guess that it's some combination of: 1) Humans making subjective calls like "generally contain claims to support" (and even just the subjectivity of the word "support"); 2) Most opinionated users (read: people who reply 1,000 times) on social media are going to "entirely, or almost entirely support one candidate"; 3) Most people are going to post things which "generally contain claims to support their arguments". But I'm not really sure if those are (to a researcher who ought to know) the obvious problems with these criteria. Is there another problem that is seen which is not listed here? (Are these suggested "problems" even problems in this case?)
- Eisenstein 3y agoAll correct. Added: you cannot establish 'ground truth' by a subjective analysis and then base your entire premise on its accuracy. They are the DOJ -- they could have found out who these people actually were if they wanted to. It isn't that hard to dox people on reddit.
- mistermann 3y agoSee also most analysis of "Russian" "trolls", etc - basically, any individual or "organization" that's not "down with the program", for which accurate information is not available, so it is hallucinated into existence.