search

wit-ai / wit

Natural Language Interface for apps and devices
https://wit.ai/
942 stars 91 forks source link

Traits added with with confidence to all utterances #2727

Open magnusburton opened 5 months ago

magnusburton commented 5 months ago

Do you want to request a feature, report a bug, or ask a question about wit? Bug

What is the current behavior? Currently, my trait missing_period is added to almost all utterances being processed. On these utterances it's displaying a confidence level of >80%. I've trained with/without it on utterances but its confidence doesn't seem to decrease as I trait utterances where it doesn't belong. Right now for the trait missing_period, I only have one value: end.

Could it be that all traits are added to each utterance and that we have to use negative trait values to train the model to "disregard" the traits? For example, on the utterances where I want the missing_period present (or reporting my only value: end), I should train it with missing_period: end. On the utterances where I don't want it to report end, I should train it as off?

What is the expected behavior? For the utterances where it doesn't belong it should not be added, or it should be added but with a low confidence score.

If applicable, what is the App ID where you are experiencing this issue? If you do not provide this, we cannot help. 602750488506714