How to import utterances into LUIS.ai

Question

The old version of the LUIS console used to have an "import utterances" function. The new console does not seem to have this function anymore.

In addition, the API (I think) used to have this option, but now it does not seem to be there.

Am I missing something, or is this a feature that is being added back-in at a later stage?

In addition, just querying the published endpoint doesn't seem to reliably 'import' the utterances into a pool. We can assume they have been imported and will eventually surface via the 'suggested' tab, but this really isn't good enough. We need to be able to import utterances in batch and then label them.

Stylite

Answer 1

The LUIS V2 API currently has an API for batch importing JSON representing the utterance and labels within the utterance which is documented here.

I'm taking it however that you're looking for a way to just upload a list of undecorated utterances and then label them in the UI?

Regarding the comment about querying the endpoint; LUIS will only show suggested utterances that it feels will help improve the model if you label them. If you label all utterances that come into the system you risk over-fitting the model to common queries.

Headset

Thanks for the response. Yes exactly we wanted to import undecorated utterances. That endpoint, and the others, require a JSON file and the file structure seems to require indicating the startPos and endPost of the entities. So we don't want to build a UI which duplicates what the LUIS UI already does - specifying the pos of the entity. I suppose we should just use the suggested tab more. The problem is that sometimes LUIS seems to need a lot of training around a set of specific utterances in order to accurately predict. So relying on the suggested tab there isn't specific enough – Stylite