We’re building a bot that should respond differently to users depending on their level of language proficiency and previous responses they provided, such as language errors made, likes expressed, etc. Rasa responses seem deterministic because they come from pre-defined templates in the utterance actions. How do we get Rasa to reply more fluidly depending on the level of the user’s language and previous extracted information from the user (errors made, content supplied)?
How to get Rasa to reply more intelligently?
I like the idea behind your question and I did some research in this area too, more into the direction of children bot mostly similar. Currently, this is something Rasa can’t provide out of the box.
A temporary workaround could be to write a
CustomAction and embedd the computational logic inside this
CustomAction such that you only need one storyline and only this Action is used. You should then choose a persistent
TrackerStore such as MondoDB such that you have access to the tracker information at any time.
Due to performance problems, it would maybe a good idea to outsource the logic and just call it inside the CustomAction.
The best way I can currently think of would be a custom Policy but as far as I know, there is currently no easy way to integrate one. Maybe we could join forces here if this topic is an important one for you.
Hi Julian. This is very helpful. It would be great to talk about this common problem. I also sent you a message at LinkedIn. I’m ready to talk to you whenever you suggest.