Symptoms
There is a Document Skill inside a Process Skill. The Online Learning option is enabled for the skill. When learning is initiated by the user in the Document Skill editor by pressing the Train button, the skill is getting trained, and fields are extracted as expected. However, if documents are processed and corrected via Manual Review, the Document Skill does not seem to get trained at all.
Cause
Online learning works a bit differently than the training by an administrator in the Document Skill:
- It does not trigger the training if the portion of the new documents is less than some threshold depending on the initial amount of samples in the document set.
- The new training model has worse quality than the previous training model, so Vantage does not apply it.
Resolution
- Process more samples. Use the following formula to calculate the approximate amount of new documents required: 1+(N*0.1) where N is the initial amount of documents in the document set before training.
- Make sure the markup is consistent.
Comments
1 comment
Adrian Enders
"It does not trigger the training if the portion of the new documents is less than some threshold depending on the initial amount of samples in the document set."
Can you expand on this comment please? What is the threshold? Or is that what the formula you mention is for? And how is this calculated for the Invoice Skills? Specifically, the Invoice US skill.
Please sign in to leave a comment.