How to fix issues/improve recognition of contracts model

Question

Hi All

I was doing some testing with the Doc Intel contracts model for a POC

below is an example that I tried for handwriting recognition (I've blurred some of the text).These are the issues:

Didn't extract contract no from title

Date was handwritten as "21st" which was extracted as "215%"

Contractor name was read as "Art Student" when it was actually "First Student"

What options do I have for correcting these and improving the recognition? All I see in the docs is to email Microsoft

Answer 1

@Mark Anderson With respect to contract number, Have you added a field during training to train your model to extract the number field from the document at the location where the number is printed or written? If you use a field and label the text that needs to be extracted, then the number should be available as value of the field.

For other issues with extraction, based on visual inspection I think the handwritten content is not totally clear and this may be the case that the OCR and Document intelligence model thinks the content to be "215" instead of "21st" for the date field and the name of contractor as "Art Student". For the name, it looks like the letters "F" and "I" are together and looks like an "A" and similarly "R" and "S" are also together and looks like a lower case "r". I would try to use a different document with the words a bit clearly distinguishable to check the accuracy of the extraction.

If you would like to provide feedback the Document intelligence portal has a smiley icon on the top right corner to provide your documents and feedback for review. Thanks!!

If this answers your query, do click Accept Answer and Yes for was this answer helpful. And, if you have any further query do let us know.