Part of the interaction model tuning process involves analyzing actual user queries coming in to ensure the model accurately matched the intent and pulled out the relevant slot information. A query can be marked as either correct, incorrect, or flagged so at a later time the incorrect matches can be corrected in the model.
The utterance match is correct, the correct information was pulled out using the correct slots, the slot values are correct.
When two people have independently marked a query as correct, it will then move to a status of confirmed correct. You will not be able to see if another user has already marked a query as correct.
Either the match was incorrect or the slots are either incomplete or incorrect.
The query requires further review, it cannot be determined if the query and matched intent is correct or incorrect.
When flagging a query as correct, incorrect, or flagged, you have the ability to add a note. This is especially helpful when just flagging the query and you want to explain why you are unable to mark it correct or incorrect. To add a note, while hovering over the flagging type, click the pencil that appears below to bring up the note field.
Article is closed for comments.