Fastbook - Chapter 18 CAM - (NLP interpretation)

One of the further research topics is :

Create a notebook like this one, but for NLP, and use it to find which words in a movie review are most significant in assessing sentiment of a particular movie review.

I have no idea how to do this but would be great to integrate with fastai library.

Closest I could find was Captum which does this for Conv and not RNN.

Does anybody know how to proceed or any examples on this topic?

fastai1 had an implementation of some text interpretation features over here:


Thank you. I’ll have a look :slight_smile: