You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
There's an edge case with this validator where it doesn't recognize products named after the competitor as a mention of the competitor.
Examples that will pass validation:
"I use an Apple Watch."
"I use microcomputers like the TRS-80 and Apple II."
It appears that the model does not recognize these as entities. So the question is really if that's the correct or expected behaviour? i.e. Is mentioning a well known product from a competitor that carries that competitor's name equivocally a mention of the competitor itself?
The text was updated successfully, but these errors were encountered:
I would speculate that the NER step/tokenization is pulling "Apple II" as a named entity and then not matching it against the known competitor 'Apple'. Our approach to the task is pretty basic: pull named entities -> check named entities against the set of competitors. (There are a few optimizations like prechecks, but that's besides the point.)
I'm not sure there's a way to handle the nuance of eponymous company products without a smarter model. I'd tried using BART MNLI and it works incredibly well for this task, but pretty much requires a GPU and might not be as fast. Not that the current model is exactly peppy.
There's an edge case with this validator where it doesn't recognize products named after the competitor as a mention of the competitor.
Examples that will pass validation:
"I use an Apple Watch."
"I use microcomputers like the TRS-80 and Apple II."
It appears that the model does not recognize these as entities. So the question is really if that's the correct or expected behaviour? i.e. Is mentioning a well known product from a competitor that carries that competitor's name equivocally a mention of the competitor itself?
The text was updated successfully, but these errors were encountered: