Re:30% of Google's Emotions Dataset Is Mislabeled
by jackbrookes on Thursday July 14, 2022 @04:30AM
(#32093496)
> Assuming I have 100k text blurbs I want to classify of sentiment and/or emotion and want at least 95% accuracy, what are the options?
The way I think Google does it with reCAPTCHA is to request classification for each item multiple times. If they differ, keep sending them out until you get a consensus on those items. It weeds out those responses that were just basically random clicks.
The way I think Google does it with reCAPTCHA is to request classification for each item multiple times. If they differ, keep sending them out until you get a consensus on those items. It weeds out those responses that were just basically random clicks.