Problem to extract data from a PDF file with multiple pages

[Roderick5h]

Hi all, 

I would like to have more ideas about the issue I'm facing with my AI model to extract custom data from documents. 

I've created an AI model to extract data from a document (in this case it's a PDF file with 4 pages, all of them are different between each other). This AI model has been trained with 4 different collections.

Each collection has more than 5 examples to help the AI model for recognizing the data I want to extract.

The issue is when I have finished the training and I do a "quick test", the uploaded file used to do the test contains an example of the same documents that I've been using for the training. The result is that the test only extracted one data, (it means one page) and it doesn't recognize the other pages.

I learnt from other publications in this community that AI builder cannot extract information from a document with multiples pages with the same structure, I just want to clarify in this case the file contains different types of documents inside and each page has different information that I want to extract.

Thank you in advance for helping me. 

[Antrod]

Hi @Roderick5h,

What I understand is that you have your 4 different "documents" (=layout of information to extract) to consider inside the same file and it's not possible for AI Builder to predict on such file.

Let's say that your model is trained to extract information of 4 documents A B C and D, each described in 4 different collections. If you extract information from a file containing A B C and D, it will mess-up with the result or just return the information of the first document.

To workaround that, you should:

• Either select the page range in the Power Automate action corresponding to one of the document from which you want to extract the information, and repeat that for the other documents inside the same file. (Page range isn't supported with the quick test for now so you won't see good results there)
• Or, create 4 different model for each of your documents and run the 4 models on your file.

AI Builder support extracting multiple page of information as long as they belong to the same document (for example an invoice with the same layout that spans over multiples pages, but not multiple different invoices in the same file).

Hope that helps!

[takolota]

You could try using an Adobe or other connector to split the files with multiple invoices into different files for each invoice & then run your model on each of those pages.

Or you could also try setting up this template & adjusting the GPT prompt to return an array of JSON objects, one object for each invoice found in the file.
https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Extract-Data-From-PDFs-and-Images-With-G...

[AlexEncodian]

@takolota / @Roderick5h 

Encodian has two split actions that could help depending on the configuration of your PDF file.

1) Split PDF by Text

Splits a PDF document based on the search text provided. You can configure how the action works including:

• Is Expression: Set whether to evaluate the 'Split Value' as a string or regular expression.
• Split Configuration: Select whether to split on the first instant, last instant or all instances pertaining to the results provided from the 'Split Value'.
• Split Action: Select a split action, whether to split on, before, after or remove the page containing the split value.

See an example solution here: Split a PDF document by Text Value or Regex using Power Automate — Encodian

2) Split PDF

Split based on number of pages (e.g. every 5 pages), array of pages (1-3,5-8, 10-12) or bookmarks. See examples in the support documentation.