I am trying to build invoice processing, while analyzing my sample forms , it gives me very few fields analyzed( required fileds are not getting identify). Is there any way to define custom fields which are not getting identified by AI builder while it gets train?
It defines the analysis on its own
Your form needs to be similar in structure and you should upload 40-50 items (ideally 100) for the model to be better in predicting and identifying patterns
If this post helps, then please consider Accept it as the solution to help the other members find it more quickly
Indeed as of today there is no posibility to define custom fields.
We can have a look and investigate why those fields are not being recognized and provide you with possible suggestions. If there is no confidential/sensitive information, would you be able to share the documents with me? You can send them to me via a private message.
I tried posting 31 documents - all are flight itinerary from one travel portal. But the results are no different than when I posted just 10 documents. The AI is missing several fields including the most important ones - the flight deails. I am happy to share the documents if that helps you troubleshoot.
Just ran into the same issues when trying to train the model with AirWayBills. the AI does not find the correct fields.
I trained the model with 10 scanned Airwaybills (as PDF) and could not load more as there is a cap at 4 MB. If I compress the PDFs I lose resolution.
As I can't show you the ones I used (as it contains confidential information) here's an example of an AWB: http://hps-trade.co.th/wp-content/uploads/2018/10/awb-format-e1539076198514.jpg
Thanks & Best regards
To give you an idea, see below picture.
E.g. the top number of the form, which has no title, is the most important number on the entire sheet, as it is an identifier number, yet the AI does not detect it.
Further, it seems to be challenged by the format as some fields are not detected at all (e.g. number of pieces in the table in the lower part of the form), or in the wrong combination (it combined gross weight, Rate class and commodity in one string) or not at all (e.g. The entire address of both shipper and consignee (all blacked out now due to privacy reasons)).
Happy to provide more feedback if required.
Congrats to the finalists of our ‘Better Together’-themed T-shirt design contest! Click for the top entries.
Fill out a quick form to claim your user group badge now!
Attend for two days of expert-led learning and innovation on topics like AI and Analytics, powered by Dynamic Communities
Features releasing from October 2019 through March 2020
Join us for the Microsoft Business Applications Virtual Launch Event on Thursday, April 2, 2020, at 8:00 AM PST.
Innovate, Collaborate, Grow - The top training and networking event across the globe for Microsoft Business Applications