Call different models according to need

Rasa Open Source
1

Hello how are you?

My system allows to execute multiple procedures. The correct thing would be that my bot works with a certain independence for each procedure. I can make each procedure use a different model example

  • process 1 ------- model 1
  • process 2 ------- model 2

is the solution I can think of to solve it, if you think there is a better I would appreciate your help

2

what do you mean use a different model example?

3

Hi! How are you?

I am with the problem that I use the chat bot in a paperwork system. the bot should answer questions independently at each step. Given the context use frequently asked questions FAQ with ResponseSelector, the problem is that there are procedures that have the same questions with different answers. so the question is how to create independence between the questions and answers of each step

Thank you very much for answering

4

I’m not sure if I understood you correctly, but I think you have at least two options:

  1. You can either create N separate chatbots for N contexts and redirrect message from user to specified chatbot.

  2. You can use only one chatbot with categorical slot set by custom action. (Details in docs: Slots). Stories example:

    story_01

    • ask_for_payments
      • action_set_centext
      • slot{“context” : “BANKING”}
      • utter_response_payments_banking

    story_02

    • ask_for_payments
      • action_set_centext
      • slot{“context” : “SALES”}
      • utter_response_payments_sales
5

Thanks for answering. Regarding point one, how can I direct the message to a specific chatbot?

On the other hand, I will also study the possible solution to point two

thanks, I wait for your answer

6

Hi,

You can train mutiple resonse selectors depending on the requirement:. The following is how it looks in config file:

 - name: ResponseSelector
    retrieval_intent: def
    epochs: 300
    random_seed: 1968
    learning_rate: 0.0006
    embedding dimension: 30

  - name: ResponseSelector
    retrieval_intent: bus
    epochs: 300
    random_seed: 1968
    learning_rate: 0.0006
    embedding_dimension: 30

I have trained 2 response selectors, one for handelling normal faqs and other for business related query. The NLU part has to be like that too, def/ml and bus/services. Hope this answers your que?

7

Hi thanks for your answer, this would be the solution. To see if I understood correctly, it would be this way, for example

image image image

If so, I would solve my problem I await your response Thank you very much