Under NLU inbox section of Rasa X GUI, we need to assign every unknown sentence entered by a user based on predicted Intent and confidence. Is there any way to make correct automatically based on certian treshold for ex if confidence>=70%, it’s marked as correct automatically without any manual intervention?
Hey Paul,
There is no way to do it according to the Rasa X API.
Closest thing to what you want is GET /api/insights/nlu/{insight_report_id}.
I would not suggest doing that though. A high confidence does not mean the bot was correct. It could be wrongly highly confident. And automatically accepting the answer will make the bot even more confident in a wrong detection.
The NLU Inbox tab exists for the bot builder to review intent and entity detections. The bot could have a high confidence in a wrong answer and a low confidence in a correct answer.
Thank you Chris.
@j.mosig and Rasa team:
-
Is there any way to let rasa mark as correct automatically if a substring is correct based on high accuracy (>=70%), for example: “English”, at the moment we have to enter in training examples all possible combinations of English word (en, eng, engl, engli, englis, english…). Can we do that without a manual intervention?
-
Do all chatbots act the same way and face this problem?
-
Is there any solution for this problem in the next releases? or do we have to keep teaching Rasa chatbot continuously?