Free text intent recognition to be extracted totally

Rasa Open Source
1

Hi there everyone! I have a problem and I hope you can show me the right way to handle it.

I defined the following snippet in the stories.md of my rasa core:

finding a person

  • find_person
    • utter_ask_person_name
  • enter_text
    • action_find_person_by_name

The last action is a custom one and it successfully extracts the text from the user response in order to make a query against a database with the text as the keyword.

The problem is that I don’t know how to define the intent enter_text inside the data.md file of the NLU in order to accept any text without falling into the fallback action.

For example I would like to have a conversation like the following:

  • (me) Find a person
  • (bot) Tell me the name of that person please
  • (me) Barack Obama
  • (bot) [action execution…] I found out Barack Obama born in 1961

Any suggestion? I can change a little the flow of the conversation, but the one I have presented would be the ideal solution :slight_smile:

Thank you in advance guys, I hope you can help me.

2

you just have to add a few different exaxmples of these things. What NLU pipeline are you using?

3

Hi! Thank you for the attention :slight_smile: I am using tensorflow as pipeline.

The point is that I would like to enter in the right intent with any word/phrase

In the example I wrote Barack Obama, but I would like to accept anything like:

  • shakira
  • how many roads
  • must a man
  • walk down
  • hello everyone
  • i am hungry
  • libertè egalitè fraternitè
  • …

I really mean any word(s) :slight_smile:

Should I restrict to the case that I can accept only people names ( full names, surnames, names, …) ? If so, can I use the tensorflow pipeline anyway? How many different example may solve the problem?

Thank you Akela for your help! :slight_smile: