When accepting user input when unknow name is given it was showing a custom fallback message how to avoid this?

Rasa Open Source
1

In my rasa chatbot when accepting user input when unknow name is given it was showing a custom fallback message how to avoid this ? how to make it apply other than name,email and mobile number?

2

Hi @skjainmiah you can implement your own fallback action and specify in your Fallback Policy configuration. For example, if you define utter_custom_fallback as you custom fallback action, you can tell your assistant to use by defining it in your policy confg as follows:

policies:
  - name: "FallbackPolicy"
    nlu_threshold: 0.3
    ambiguity_threshold: 0.1
    core_threshold: 0.3
    fallback_action_name: 'utter_custom_fallback'
3

Hi @Juste, I already have my custom “FallbackPolicy” but the thing is I’m using forms to collect data from user like name, mobile and email then when the user enter the name it shows Fallbackpolicy message which I set. Actually it should accept user input but it doesn’t because it hasn’t trained the name entered by user. So how to solve this problem?

It means forms and fallback policies both cannot be implemented at a time am I right?

4

Hi @skjainmiah. Thank you so much for clarifying your question :slight_smile:

You should be able to implement forms and fallback policy in the same bot. If the assistants responds with fallback to most of the user inputs there are a few things I would check:

  1. How your NLU model is performing. Are intents classified as expected? You can see what your assistant does at every step of the conversations by running it in a debug mode: rasa shell --debug. If the intents are classified incorrectly I would go back to improving the NLU data until the model performs better.

  2. How your fallback policy is configured. I would look into what nlu_threshold and core_threshold you have set. If you set it to close to 1, there is very little room for error that you give your assistant meaning that even if the assistant classified the intent correctly but with a lower confidence, your assistant will respond with a fallback.

Would you be able to share the files of your assistant. I could try to replicate it and give you more advice on how to proceed.