For NLU training, is there any way to identify “undefined entity”?
For example, in nlu.md an intent is defined such as
## intent: Inquery_Details
- What’s the exact [foobar](detail)
For an input sentence like:
“What’s the exact address”
Is there a way for Rasa to not only identify it as the Inquery_Detail intent, but also recognize “address” as a detail entity? I am trying to put a “variable” in the training sentence and extract entities that may not have been defined in the training file.
Hi @eugenelin89. In your training data you would need at least a few examples with actual entities values - for example address.
One more thing to keep in mind - if you label the entities as detail your model will not be able to classify those entities based on what they actually are (addresses, numbers, etc).
I gave it a few more additional examples so the definition of my intent looks like:
## intent: Inquery_Details
- What's the exact [address](detail)
- What's the exact [location](detail)
- What's the exact [item](detail)
- What's the exact [answer](detail)
- What's the exact [number](detail)
But in the rasa shell when I type the sentence:
“What’s the exact message”
Though Rasa recognize the sentence to be Inquery_Details intent, it still does not recognize the word “message” as a detail entity.
I am just wondering if I understood your earlier suggestion correctly?