Validating user data

Say, I’ve a RASA chatbot running on messenger. I’ve install RASA on GCP VM using docker. The question is, When users interact with my chatbot how can I train those interactions? Thanks in advance for your help :slightly_smiling_face:.

If you have Rasa X installed, you can easily see the user messages and their predicted intents in the NLU Inbox tab.

If the detected intent is correct, you can click on :white_check_mark: to add it to the NLU data to reinforce this behavior. If not, you can correct it and click on :white_check_mark: to add it to the NLU data to correct this behavior.

This is useful because you can sort and filter messages by confidence (among others).

You can also directly annotate the NLU examples in the Conversation Inbox tab if you come across an error there.

This is useful because you can sort and filter conversations by whether or not an unhappy intent or bad feedback slot was detected (among others).


Adding to the @ChrisRahme suggestion, if you are even working on rasa open source, try add the image of rasa x docker image simultaneously, what ever chat you are doing with rasa open source, you will be able to see in NLU Inbox tab and even the training examples or even out of scope questions. Chris given you best use case example :clap:, how you can validate data at the backend. Just you need to focus on confidence score for the out of scope questions.


Thank you for your reply @ChrisRahme . Another question according to your answer, If I correct the incoming massages using rasa x, it will save as example of intent right? At the same time I’m developing my project on GitHub. Then how they sync each other?

1 Like

That’s right

Rasa X has Integrated Version Control :slight_smile: (tutorial)

I see. Thank you very much :smiling_face_with_three_hearts:.

1 Like

Happy to help :slight_smile: Please mark the answer as solution if you think it is useful.

1 Like