@th3j35t3r
Weak writing all around. How did they select sites, messages?
An article about rates that doesn't include volume is just junk. The "study" they cite is the same. "Detected 325,739 tweets" _out of how many examined?_ We don't know. Randomly selected? Maybe.
Did the comments become super-saturated? Encoded messages inside the messages? "Gab, now 150% anti-Jew"
"bespoke speech classifier" made by "combining over twenty leading machine-learning models":= 🤮 Black box.
I'll stop.
@b4cks4w @th3j35t3r "Antisemitism can be quite difficult to spot because of its more covert nature,” she says. In other words, their bespoke speech classifier might also misclassify support for Palestinian civilians as antisemitic speech.