question

screamingbuddha avatar image
0 Votes"
screamingbuddha asked romungi-MSFT commented

Form Recognizer - Composing models with differnt entities

@romungi-MSFT If there are models with different entities to be extracted. Is it recommended to compose those models or it it's better to have it separated? The reason is because we can avoid the document to classify and map to the model.

azure-form-recognizer
· 3
5 |1600 characters needed characters left characters exceeded

Up to 10 attachments (including images) can be used with a maximum of 3.0 MiB each and 30.0 MiB total.

@screamingbuddha Do you mean extracting different fields of a single document? If Yes, then using different models to scan a single document means multiple calls to the API and processing the response of these calls for your application. Composing models helps you reduce this and you can always assign friendly names to individual model or composed model to compare the results later.

1 Vote 1 ·

@romungi-MSFT Thanks for the quick response. Actually, I have two sets of documents with different entities. I would probably have to create a separate models but the prediction happens with a single source. In this case, I'll probably have to create a logic outside of FR to classify the document and map it to the corresponding model. Or is there a better approach to it using Form Recognizer?

0 Votes 0 ·

@screamingbuddha I would try to check the results with a composed model with couple of entities. If this does not provide satisfactory results you can use the individual model and collate the results for the document in your application process. If the management becomes complex you could as well create another FR resource and create models for a specific scenario but you might have to also manage different endpoint and keys.

I think the first option should work if the training data covers all the fields that you want to extract.

1 Vote 1 ·

0 Answers