Entity recognition in similar training phrases

Rasa Open Source
1

I am designing a bot for a navigation service.

The problem is that users can use the same phrases with different entities. For example, “Find a cafe” and “find a KFC”.

nlu:
- intent: searchPlaces
  examples: |
    - Find a [food]{"entity":"placeCategory"}
    - Find a [Cafe]{"entity":"placeType"}
    - Find [KFC]{"entity":"placeName"}

stories:
- story: search/main
  steps:
  - intent: searchPlaces
  - or:
    - slot_was_set:
      - PlaceName
    - slot_was_set:
      - PlaceType
    - slot_was_set:
      - PlaceCategory
  - action: utter_search

responses:
  utter_search:
    - condition:
        - type: slot
          name: PlaceName
          value: 
      text: "A list of places called {placeName}"
    - condition:
        - type: slot
          name: PlaceType
          value: 
      text: "A list of places with type {placeType}"
    - condition:
        - type: slot
          name: PlaceCategory
          value: 
      text: "A list of places with category {placeCategory}"
    - text: "Nothing found"

As a result, the bot chooses a random response, even if it does not contain the correct entity. Sometimes the bot substitutes entities from previous requests, sometimes it simply replaces the value with “None”.

How can I fix it?