For this version we used TextBlob out of the box with some custom weighting to avoid false positives. We'll relaunch in about two weeks with an improved model

For this version we used TextBlob out of the box with some custom weighting to avoid false positives. We'll relaunch in about two weeks with an improved model that we're currently training on a much larger labeled dataset and SOTA BERT model. :)