How to support "wildcard" entities?

Rasa Open Source
1

Let’s say I have an application that has a database of products, and the user may do operations on them (for instance, set the price). I want to let the user do these operations using a Rasa Chatbot.

What I’m trying to do, is have training scenarios with the following structure:

# domain.yml
entities:
- name
slots:
  name:
    type: text
    mappings:
    - type: from_entity
      entity: name

#nlu.yml
- intent: set_price
  examples: |
    - set price of product [IPhone X](name)

The problem here is that the name of the product is completely arbitrary. The user creates the product, so the name can be anything. Is it possible to still train Rasa in order to extract the product name? I’m not sure if using spaCy extractors like for Person names or locations works here because the names are random…

Right now, the solution I’m using is having the name slot with mapping from_text, therefore the dialog flow would be:

user: Set price of product

bot: what is the product?

user: IPhone

This works, but it would be ideal to do this in one sentence…

2

I have another question. I’m still using the flow I mentioned using the from_text mapping, however I have a problem. When the user answers with the name of the product, the slot gets filled, but sometimes it also fills other entities which actually already had a value. Is there any way to prevent other entities from being overwritten when the user is asked for the name? (I am using the default pipeline with the DIET classifier)