How should we add our data to the file “nlu.md” so that it is optimized and gives the best accuracy for the model? For example, if I add 10 types for each question, is it enough?
The second:
I need rasa answers to 10 types of project questions, for example answering health care questions, answering cooking stuff, …
Do I need to learn one rasa project with all the questions and answers or is it better to use a separate rasa project for each project?
note that each project might contain more than 50 questions and 50 actions.
It’s better to utilize the same rasa model for all questions since they are distinguishable from their phrases. However, it is possible to maintain separate nlu.md files that let you separate your phrases for organization purposes, but they all get used towards training a single model.
I would add the training data based on all the different ways you can say a particular phrase for something. So think of all the different ways the user may ask for the same thing and add them as your training data for each one. I personally don’t have too many duplicate phrases. If I do, it’s to train the model for a different entity that’s not being detected well. The thing I liked about Rasa was that it didn’t require too many samples to train it. Training it with 1 phrase ensured that particular phrase would be predicted as the intent it was under.