DialogFlow - does using entity for verbs and adjectives degrade model performance

26 views Asked by At

I'm hoping to get some clarification around the use of entity in DialogFlow ES/CX.

We currently use custom entities heavily in our ES instance to handle variety of user inputs and help with context recognition, from:

  • nouns e.g., product type, plan type

  • verbs e.g., entity @buy to include verbs expressing users wanting to purchase i.e. "buy", "purchase", "looking for" etc OR entity @cancel to include verbs expressing users wanting to cancel the order i.e. "cancel", "stop", "terminate" etc

  • adjectives e.g., entity to capture users negative sentiments i.e., "upset", "disappointed", "angry" etc

On Google documentation I can only see entities being used for names, types, dates, numbers etc. and there is no mention or advice on the pros and cons of using entities outside of these examples. I'm trying to understand if the way we currently set up entities may have any unwanted impact on the ML performance as we are seeing some issues with the model whereas an utterance with just slight difference to a training phrase is being unmatched despite having a lot of training phrases and entities.

0

There are 0 answers