About utter_default

Rasa Open Source
1

I have defined around 30 intents and trained the model and made many interactive sessions. But, when I’m trying to ask the questions to the bot which are not related to the intents It’s not performing utter_deafult action. It is showing intents which we defined for the questions which are not at all related to that intent or it is showing none. Can you please help me overcome this issue. Thanks in advance.

1 Like
2

You will need to visit your fallback policy’s setting and check your threshold.

In my case it’s:

  - name: "FallbackPolicy"
    nlu_threshold: 0.5
    core_threshold: 0.5
    fallback_action_name: "action_default_fallback"

When my confidence level is below 0.5, the fallback will then be triggered.

Also, you can run rasa shell nlu

to check your intent ranking and scorings as a way to test

3 Likes
3

Thanks for the reply. What should I do when I ask the bot a question and it returned none?

4

This depends on your design for in your stories. Ideally for every intent that gives an utter reply, you will need to write a 2 liner stories for these cases

## Story 1 
* intent_name1
 - utter_response1 

## Story 2 
* intent_name2
 - utter_response2

then for complicated logic flow, you will need to design the longer stories accordingly

5

I did not get you. Can you explain in detail please

6

@mohankumar can you post an example input where the intent is recognized as None? Also, a good way to avoid It is showing intents which we defined for the questions which are not at all related to that intent is to create another intent such as out_of_scope. In this new intent you can add all the data that the bot can’t handle. Then it is more likely to classify any unsupported intents as out_of_scope instead of as a supported intent. Then you can also directly invoke the default response in the stories:

## Story 3 
* out_of_scope
 - action_default_fallback
7

Thanks for the reply. here is the example

## intent:cashless_process_flow
- can you tell me the [cashless process flow](information)?
- what is the process flow for [cashless hospitilization](information)?

## lookup:information
- cashless process flow
- cashless hospitilization

entities:
- information

slots:
information:
type: text

i defined all the things,but it not identifying the entities,can u plz help me

8

Is this all of your training data for that intent/entity? Unfortunately it’s probably not enough to pick up the entities – you’ll want more examples.

9

nlu.md

intent:health_insurance

-what is health insurance

stories.md:

*health_insurance

  • utter_health_insurance

In the above defined intent if I ask bot the question saying “what is health insurance” it will identify the health_insurance intent perfectly. But, when I ask the bot a question saying “what is life insurance” then also it is identifying the health_insurance intent. So, this is invalid because the question is not valid regarding to intent.

when I ask the question saying what is health insurance the confidence level is 97% and when I ask the question saying what is life insurance the confidence level is 80-90% So, this is wrong in my case because it should utter default response.

So, how should I approach my model like with only intents or intents and entities because I’m confused with the confidence levels of intents and entities.So, I want to know if response is given based on only intent’s confidence or the confidence level of both intents and entities.

As of now In my model I have used only intents and not entities

please help with this issue. Thanks in advance.