Rasa followup questions using fallback classifier

Hi! I want to use a customfallback classifier where instead of utter_rephrase, I want the bot to ask follow up questions. As an example, I have an intent for ‘drink’ and assuming the user enters an entity which the system did not extract, the bot will predict the intent ‘drink’ but due to a low confidence, the fallback would be initiated. In the fallback, I want to enable a conersation flow as follows:

Story 1:

–Can I have espresso?

Is that a drink?


Can you confirm the drink?

–Espresso (internallly set slot)

Let me get you espresso

Story 2:

–Can I have espresso?

Is that a drink? –No . . .

Further follow up questions

Sorry I cant help you

I have read through custom actions, forms, branching but cant find the best way to implement this. Would like to get any inputs and ideas on implementing this.

I think that custom actions are the way to go here, but it depends a bit on what is “normal behavior” in the assistant. You can have a custom action that generates buttons that the user might press and you can have the same custom action check if there is an entity in the current utterance.

What is the goal of the assistant?

@koaning We are testing a chatbot for a care robot which can achieve basic task oriented conversation. We use the spacy package to try covering the entities along with training data. But for cases where the entities are not recogonised, we want to customise the fallback to handle unknown instances by asking basic followup questions. To make it more clear, let’s say espresso is not identified as an entity, but the bot asks if its a drink and if the user confirms it, it is manually set as an entity and intent gets identified as drink. I understand using interactive learning and capturing conversations might be the best way to go about it. I have tried to use different form and custom action logic to implement this but not with much success. I’ll try to implement buttons to do this and see if they give the required results. Do let me know if you have any further inputs or require more clarification

1 Like