Closed guillegette closed 7 years ago
This would be useful, because it could enable handling of fallback intents, which API.AI has but I assume aren't included/possible in Rasa as it is currently (is that right?)
Yes that's right. What does work is to train a generic intent like none
or outofscope
. I could see how a confidence value would be really nice to have though.
@amn41 Any thoughts on how to train a good none
intent, like the one that comes with Luis.
@designerkamal In my experience it works quite well to just have a bunch of "out of scope" sentences labelled as None
. This seems to generalise quite well. If you are having trouble please feel free to email me and I will think of ways you can improve perf :)
awesome!!
what is the logic behind deciding the confidence value? Is there any threshold? I want to know this for designing my chatbot
Would be great to return the confidence for an intent and/or entity, how does it work at the moment? does it return the best match ? Happy to help implementing this as well :)