Model to NLU data conversion


The model we use in our project stands as follows :

  • Annotated sentences for intents and entity training (eg. I want the yellow color)
  • List of values (may not be comprehensive) for an entity (eg. color -> [blue, yellow, red, pink])
    • List possible “sub-values” for each of the possibilities (eg. blue -> [ultramarine, navy])

Here is how I see this model translated into RASA NLU data :

- I want the [yellow](color) color

- blue
- yellow
- red
- pink

- ultramarine
- navy

What do you think about that ? Any idea, recommendation (I can not change the model) ?

Thank you, Pierre

I am assuming that you are talking about training data’s format

Let me share with you my Weatherbot’s location lookup.

## intent:Function_Inform_Location
- What's the weather in [Berlin](location) at the moment?
- What's the weather today?
- What's the weather in [London](location) today?
- Show me what's the weather in [Paris](location)
- I wonder what is the weather in [Vilnius](location) right now?
- what is the weather?
- Tell me the weather
- Is the weather nice in [Barcelona](location) today?
- I am going to [London](location) today and I wonder what is the weather out there?
- I am planning my trip to [Amsterdam](location). What is the weather out there?
- Show me the weather in [Dublin](location)?
- [Beijing](location)
- [Tokyo](location)

## lookup:location

For the lookup table, it’s just a text file stored in the inner lookuptable folder. I have around 120,000 city name in it. It’s easier to train it that way and same much training time if I were to map all the cities into individual annotated entities in my

For synonym. you may use the ultramarine way of mapping it. However, I would recommend you to design a custom action to iterate through another list of color subsets if you want to have a very consistent result.

You may refer to the offical doc here too

1 Like