Interactive learning for NLU

Hello community ! I have a question , so we all know that interactive learning is used for rasa core part to train the model for bot’s action , but is there another use for interactive learning for NLU part , to train the model in order to knows the relevant entities in a message and therefore increase their confidence indices ?

Hey @anojoual. With the interactive learning you can save all new user inputs and use them to improve your assistant (you can also annotate new entites and incorporate them to your training examples much easier)

Hello, I have a question related to this issue: when using the interactive learning mode it asks you to save the new NLU data in your markdown training file. How can I save that data if I’m using different json files to train the model?


Hi @juliamendoim. Sorry for a late response. You can actually convert json to md files quite easily with Rasa. For example the command below will convert json to md:

python -m rasa_nlu.convert --data_file=Emploji.json --o --language=de --format=md

So you can save interactive learning data as md and the convert to json easily and add to your original training examples.

Thank you @Juste. I actually have both types of files in my training data folder and it seems to be working fine.

hello @juste how to you create stories for forms ? When I tried interactive learning, there is no form action coming up for me to “train” the dialogue. After I key in the first slot value, it just shows utter_slot_values but I want to train each form prompt for the form slots.