Hi,
I am trying to explain predictions from a fine-tuned bert model for a binary text classification problem using LIME.
LIME allows us to extract the top n features or words most relevant in the sentence. I was wondering if we can change the LIME explainer in some way to obtain both bi-gram and uni-gram features/words?
And other than LIME and SHAP, are there any similar explainers available to understand the predictions of a BERT model better?
Thank you so much!!!