Annotate large corpus / Import it into Rasa X

Hi, I’m trying to bootstrap an agent, I have data from real users conversations to a human agent (~12,000 conversations), and I’d need to import it to Rasa X to start annotating it both on the user and the agent side, but as far as I’ve seen, the APIs don’t allow me to import untagged conversations so that I can annotate it from the UI. Or am I missing something?


Hi @kappa90 you can import unannotated user messages, but not full unannotated user conversations. The conversation flow is something you should design yourself first, and then once you have a first minimum viable assistant, you can have users interact with the bot and annotate that data.

Hi @akelad,

What you are saying is that:

we cannot automatically get/annotate “stories” from existing conversations? First we need to manually design some stories and build a MV bot, then again manually start interacting with this bot to collect more data and improve the bot?

For me, building a MV bot using manual design completely makes sense but after that we should have a kind of automatic mechanism to load the existing conversations in Rasa/Rasa X instead of manually interacting and entering the conversations we already collected.

Does Rasa/Rasa X has such a capability? Any thoughts on that? Thanks,


The issue is they’re completely unlabeled. And I would guess your bot is going to behave a bit differently to how a human agent behaves, so it doesn’t make sense to import those conversations because the flow will probably be different