Form Recognizer handle multiple of W2 forms in single pdf

Harshil K Kansagara 1

Hello,

I have a requirements of processing W2 files with Form Recognizer. The number of W2 Forms per single PDF files may vary from 1 to n number having same layout in each page and per page there will be 1 employee W2 form.

I can't used Prebuilt W2 Model which is provided by Microsoft Form Recognizer because we are consuming different key fields name in application. I'm using Custom Extraction Model to label the W2 files which consists of 1 page pdf and training the model using using neural build model.

When I'm trying to upload more than 1 page pdf document for testing purpose, I'm not getting output from each and every pages. But if I upload same document for the testing purpose in Prebuilt W2 Model, I'm able to extract each key fields value from each pages in given pdf document separated out by pagenumber in response.

Could you please suggest how to handle such scenarios?

How can I get per page extracted fields in the output response?
What training mechanism is using in generating Prebuilt W2 model provided by Microsoft Form Recognizer?

Please let me know if any more information is required.

Thank you.

Harshil Kansagara

YutongTie-MSFT 48,586 Reputation points

2023-05-04T19:50:27.8933333+00:00

Hello @Harshil K Kansagara Thanks for reaching out to us, which tier you are in? Unfortunately, with a free tier subscription, only the first two pages are processed, this may cause your issue.

Regarding the training mechanism used in generating the prebuilt W2 model provided by Microsoft Form Recognizer, it uses a combination of supervised learning and neural network-based models. The model is trained on a large dataset of W2 forms, with a focus on identifying and extracting key fields such as the employee's name, social security number, and wage information. Microsoft has not released details about the specific architecture or training methodology used in their prebuilt W2 model, as this is considered proprietary information.

Regards,

Yutong
Harshil K Kansagara 1 Reputation point

2023-05-05T03:30:59.83+00:00

Hi @YutongTie-MSFT , Thanks for the quick response. I'm using S0 standard pricing tier form recognizer account.

My question is while analyzing a document it is getting completely analyzed but in the JSON output it is only picking first page key value pairs. So why it is not picking key value pairs from another pages even though the data in other pages not similar to data in first pages?

Regards,
Harshil

Share via

Form Recognizer handle multiple of W2 forms in single pdf