The perspectives toxicity API is a severely flawed outcome that shouldn’t ship. Not in this state.https://twitter.com/violetblue/status/903697800226979840 …
If you train with the right corpora, yes. The problem is they train with BS designed to be "neutral". "Civility" vs "incivility".
-
-
Train with 4ch*n+8ch*n+hate-themed subreddits+g*b as your corpus, wide variety of nonabusive forums as control, and it'll be 100% accurate.
-
I doubt it'd be able to handle context-dependent stuff (like sarcasm) effectively, still
-
text is the wrong thing to be looking at, you want embedding vectors from the social graph
-
I should clarify: 100% accurate at categorizing a user given access to all their posts. Not necessarily at classifying one post.
-
only for casual harassment, maybe. definitely not for organized griefing
-
if you create an account for the sole purpose of greifing you can control the context
End of conversation
New conversation -
-
-
and also apparently they just use a linear classifier over bag-of-words vectors and call it good?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.