Intents not correctly classifying/overfitting[maybe]

Rasa Open Source
1

I am having two intents, one is x and other is y. In both intents, I am having >~20 utterances and in x intent, a specific word ‘z’ is repeating a lot as compared in y. When I send the query with word z in it, it classifies it in x intent but it should be classified in y. How can we overcome this? Is inform intent solution to this because entities are swapping in this case. E.g: x -> a and b are friends y -> b and a are not friends

It’s almost like this. What can we do in such situation?

PS: I also want to learn how to create inform intent. Should we have to use only inform or inform_<name> will do?

2

If just the entities are making the difference, then it seems indeed like a case for an inform intent :+1: In case of the negation it indeed seems more like some overfitting. Try providing more data and also check if your data is clean.

I also want to learn how to create inform intent. Should we have to use only inform or inform_<name> will do?

One single inform intent is enough.

3

Can you please give a small example? It will be really helpful.

Thanks

4

Are these examples good rasa-demo/nlu.md at master · RasaHQ/rasa-demo · GitHub ? They are from the assistant on our docs Build contextual chatbots and AI assistants with Rasa