One more idea. You could add a "system error button" or something similar that a user can click and tell you how your NLP/ML algorithms work. Then, use it for i

One more idea. You could add a "system error button" or something similar that a user can click and tell you how your NLP/ML algorithms work. Then, use it for improving your models or maybe even implement active learning.