I tried putting blank in the intent field in md file but it isn’t present in the
training_data.json file after model trains so I assume that the data without intent is not read.
- some data goes here
- another data goes here
So why do you want this exactly?
It’s to train the model on words that can come but don’t want to put it in the intent examples as it would overfit.
This is needed when entities have many synonyms and putting all of them with the same data format only leads to model overfitting the data.
You can put it in separate out_of_scope intents. This is a sort of trash bag intent for out of scope messages.
I’ve got specific (synonyms of) entities in those examples so I don’t want it to mispredict it as an
out_of_scope_intent. I just want the model to know that those words exist.
Then it sounds to me that those examples are needed to train the model anyways. I get your concern about overfitting, but as long as you capture the intents you have in different natural language examples, you don’t really have to worry about overfitting that much.
I’ll just consider trying to move data to json format.
I don’t think having that many examples with same format is worth it for the synonyms of a single entity value of a single entity type.