In Microsoft Luis, how do I model an array of entities?

Question

So I am trying to understand user's utterance for selecting an array of entities, here are some example utterances:

My choices are 1, 2, 3, and 4
Let's go with Red  Blue  Yellow
How about January and February and March
I want A & B & C

So in the example utterances above, I should be able to get the following entities:

[1,2,3,4]
[Red,Blue,Yellow]
[January,February,March]
[A,B,C]

I don't have the choices before hand so I can't train them. But I want to be able to understand multiple choices that the user select. Is this possible within Luis?

I guess I could mimic NLU by just use delimiter and regex to break up the user's input. But that wont work if the user use different delimiter. Or if the user append a lot of chatter before hand.

Answer 1

You can solve that by creating an Entity of type Simple called Choice and adding utterances like that and labeling the choices with said entity, like this:

Luis, as an AI, is very good at recognizing patterns and those utterances have a very definite pattern which is:

Utterance: Prefix + Choice1 + delimiter + Choice2 + delimiter + ... + ChoiceN

Being

• Prefix: "My choices are", "Let's go with", "How about", etc.
• Choice: any word
• delimiter: ",", "and", " ", etc

That is roughly the pattern that Luis detects, in this case will detect anything after the prefix separated by a delimiter as a choice. So an utterance like

I want a soda, burger and fries

Will result in:

Even though we haven't specified Luis that soda burger and fries are valid choices, he still detected them as such!

Here's the Luis model I created for this answer: https://github.com/navelDirt/luis-apps/blob/master/ChoiceDetect/ChoicesApp.json

You can import it in Luis by going to YourApp -> Manage -> Versions -> Import Version

Edit:

It should detect N choices