Is there a better way to handle multiple intents in comparision to the way in it is currently handled?


(Ankita) #1

I am using rasa to build a chatbot and trying to explore its capabilities as a tool. Rasa handles multiple intents by training the nlu with multi-intent examples (as the one given below)

intent: affirm+ask_transport

  • Yes. How do I get there?
  • Sounds good. Do you know how I could get there from home?

So, in case when you have more than 5 to 6 intents and all their possible combinations, then it becomes hectic to provide all the required data to train nlu.

In such scenarios, is there a better way to train nlu for multiple intents

(Alan Nichol) #2

Hi Ankita,

In theory, yes you would have a large number of combined intents this way. However, in practice you will only ever see a few of these coming from your users. I would strongly recommend only including those in your training data, and not trying to create artificial training data for every possible combination.

(Ankita) #3

Thanks Alan,

But I really have a situation in which we need scaling around this. It wont be possible with this alternative.

(Alan Nichol) #4

that’s interesting. Can you please explain the situation? maybe there’s another way