RASA Dialog Management: Ho to migrate state-based approach to a story-based approach?

335 views Asked by At

RASA NLU (intents and entities classification) is clear, simple and well implemented.

But one of the things makes RASA different from many competitors is RASA dialog machine-learning based dialog management (using stories and rules).

Myself (and I guess many of people that use non RASA framework) I used to manage task-oriented (conversational) workflows using the concept of state-machine. BTW I developed a small open source project NaifJs, that conceptualize a conversation as a sequence (a graph) of states, where each state is a set of valid intents involved in a conversation context.

https://github.com/solyarisoftware/naifjs/blob/master/doc/img/state-machine-1-new.png

So, what is not clear to me is ho to migrate a state-based (or context-based) approach to a story-based approach. Just to simplify/exemplify, I could immagine that given a " dialog flow" realized by conditional statements, I have to "serialize/duplicate" each "if" in multiple "then"stories.

Do you have already explained this problem in some articles? Maybe a gift for a new RASA blog article/youtube workshop?


Thanks

Giorgio

0

There are 0 answers