Are entity-only training examples still supported? If so, how are they formatted?

The docs for the training data format mentions both the intent and entity are optional data, text being the only required data. I’m just not sure how examples that include the text and entities, but no intent should actually look. In older docs I see mention of a entity_examples, but that seems to have been removed from the more modern docs.

I’m interested in this because I would like to build a lot of training examples for the CRFEntityExtractor extractor without introducing a large imbalance in the intent training data.

Hi @plwrenn You can use the json data format to accomplish this. Here is an example that should work -

{
  "rasa_nlu_data": {
    ...
    "common_examples": [
      {
        "text": "hey",
        "intent": "greet", 
        "entities": []
      }, 
      {
        "text": "I am looking for asian fusion food",
        "entities": [
          {
            "start": 17,
            "end": 29,
            "value": "asian fusion",
            "entity": "cuisine"
          }
        ]
      },
      .... 

Let me know if you have any further queries.

1 Like