Is it possible to use some more attributes other than text and response for a message and featurize them for both training and interpretation? So basically, instead of having only text and label in the training data, is it possible to have text, context-1, contetxt-2 along with label?

Thanks in advance!

Hi @yuga04 . What kind of other attributes do you want to use? By context-1, context-2 do you mean conversation contexts? If yes, then that wouldn’t be possible because you don’t have those during training. However, you may be interested in looking into end to end training scheme.

Hi @dakshvar22. Basically instead of just text(string), I wanted to encode & embed some more features for intent classification. Currently, If I don’t use entities in the training data or the pipeline, only the text is featurized and used for training. I was wondering if there is a mechanism to add some more features for the training along with the text.

PS: This is with respect to only intent classification. context-1 and context-2 are not conversational contexts, rather some information defined by me to use it in the sequence classification training.