Decrease confidence without out of scope

Hello everyone, I am doing a simple rasa bot. My domain config is default one, i kept intent unhappy, affirm and deny. When i talk to my bot and say a simple good morning i get an intent confidence of 1.0 on unhappy, i would like to have that score way lower without adding out of scope intent.


Hi @enamort. I would recommend collecting more training data for your intents to build a better NLU model. If you want to handle a simple good morning, then I would also suggest adding a greet intent

the idea behind that is simply to have a low confidence on a sentence far from the intent. I tried linear norm and cross entropy on a DietClassifier but the result is the same.

@enamort ever find a solution? I am facing the same problem when trying to trigger a fallback function.