Customer question: how do we know which words/n-grams increase confidence in the predictions?
The word cloud is useful for this!
You can also look at the coefficients directly for any linear model with ngrams.
Finally we’ve got Prediction Explanations for text features.
Thanks for the response. I love these and they make sense. I recommended word cloud but she indicated that that was intensity not confidence (to me they are highly related).
Our linear models are regularized so low confidence words will be dropped.