Get back user message without NLU interpretation on it

Hi community,

I have the need to send feedback from the user about its experience on the bot to my backoffice. I wonder whether it would be possible to:

1/ When the bot asks for user feedback → let the user write its feedback on natural language through the bot 2/ Not interpret this user feedback with NLU (not try to extract intent/entities) 3/ Just retrieve this user feedback in custom Action via the Tracker to send it to my backlog

Thanks for helping !

1 Like

Hi @mattvan83, what you can do is create a Feedback Form that uses the from_text mapping to retrieve the user’s feedback message and use a FormValidation method to send it to your backlog.

Hi @b-quachtran thanks for the advice.

How would it be possible however not to interpret the user message from NLU?

@mattvan83 Does it fit your use case if the message is parsed by NLU but the intents / entities extracted are not factored into what the dialogue manager does with that information?

@b-quachtran Yes it would feed my use case if the message is parsed by NLU but the intents / entities extracted are not factored into what the dialogue manager does with that information.

  1. How would you do this?
  2. If the intents / entities extracted are not factored into what the dialogue manager does with that information, how do you manage the dialogue then?
  3. I am already in a Form, so how could I manage this Feedback Form within the current Form?

@mattvan83 This can be done by implementing a Form and using the from_text mapping. You’ll need to define a new slot to store the user’s feedback information and the from_text mapping will store the user’s input into that slot regardless of the parsed NLU data:

responses:
  utter_ask_feedback:
  - text: What feedback do you have about the user experience?

slots:
  feedback:
    type: any

my_form:
  ...
  feedback:
  - type: from_text
1 Like

@b-quachtran thanks for the details.

It indeed works ! However, could you explain to me why when using:

my_form:
  ...
  feedback:
  - type: from_text
    intent: None

it doesn’t work and is stucked in searching intent not defined in domain file, while without it seems that NLU interpreter doesn’t search after intent?