Understanding Form Recognizer Composed Model Classification Step

Question

Hi Team,

While looking throughout composed model documentation I found a lot of helpful information. However, there are no details on classification step performed by composed model. Can you give me some insights what exactly happens in that step? Based on what metric given model is selected? What is taken into consideration for classification algorithm?

Composed model seems to be a perfect solution in my case but it sometimes mismatches model with document. Before going further with that solution I would like to understand it more.

Thanks!

Answer 1

Hello @Tymon Lewandowski , Thanks for using Microsoft Q&A Platform.

To better understand Composed Models, it may be helpful to first understand the Classification Model and the differences between the two models.

Custom classification models are a type of deep-learning model that use both layout and language features to accurately detect and identify documents within your application. These models classify an input file one page at a time to identify the document(s) within. This model is useful in scenarios especially to identify the document type before extracting the information.

By pairing a classification model with a custom extraction model, you can analyze and extract fields from forms and documents that are specific to your business, creating a document processing solution. Additionally, standalone custom extraction models can be combined to create composed models. When a document is submitted for analysis using a composed model, the service performs a classification process to determine which custom model is the best fit for the submitted document.

Document Intelligence uses advanced machine learning technology to identify documents, detect and extract information from forms and documents, and return the extracted data in a structured JSON output.

I hope this helps.

Regards,
Vasavi

-Please kindly accept the answer and vote 'yes' if you feel helpful to support the community, thanks.