Quote:
Originally Posted by erimir
Quote:
Originally Posted by Kamilah Hauptmann
|
I am familiar with some of the algorithms used therein.
I wonder how different the results would be if using measures based on the content of posts rather than the identity of the posters.
Someone with stronger modeling skills than me would be able to come up with a way to combine the two aspects somehow.
I've done some stuff with topic modeling, and you can get sensible results from combining topics in that case as well.
For example (the posts here came from a board for gang members, hence it was male dominated):
Posts scoring high in both the race relations + women/dating topics gave you posts about... interracial dating. You could get similar results by combining topics focused on rap/music, religion, particular gangs, etc.
Another interesting tidbit from that forum was that I found over 200 spellings of motherfucker(s). |
You might find the Reddit discussing the results interesting:
Dissecting Trump's Most Rabid Online Following : dataisbeautiful
The guy who ran the numbers is (was?) in there too, responding to questions.