repeated field on the same page

[lavez]

hi; thank you for all, these are very exciting features.

I've a special case; one field that I want to capture that is repeated on the same page. apparently there is no way to manage this because when it find the first time the searched field then it go to the next page. could you help me on this? Many thanks

[JoeF-MSFT · Accepted solution]

Hi @lavez,

Thanks for asking! 

In this case, does the field always have the same value or different values?

If it has different values, what I would then recommend is to define as many fields as there will be in the page. If the field can be present up to 5 times in the page, define 5 fields and tag those during training.  

If the fields always have the same value - I'd love to learn more why you would want the AI model to extract it multiple times.

Let me know if this is not clear. Happy end of year holidays! ☃️

[lavez]

Hi! thank you

it's a transactions confirmation report....it's not organized as a table but there are two or three shares per page...

I will try your idea, I think it could works

[Hiwm]

Hi,

I have an important question I need help can help with:

I have similar fields with different values in my documents. will AI detect all? as i have no idea how many would be in each document. EX: one document can have 1 customer, other document can have 5 customers and so on...so can I train the model with one example per field per file?

Please advise,

Thank you!

[Antrod]

Hi @Hiwm ,

It would depend on how the customers are displayed in your document. I'd say that one possible solution could be to declare as many fields as you could have customers. For example, if you know that you can have maximum 5 customers, you need to declare 5 customer fields. Then in your training set, make sure you have examples of each number of customers. This is hypothetical as it would depend on how the layout varies when there are different number of customers, but that a solution to explore.

[Hiwm]

Thanks @Antrod ,

I should declare as many fields possible per document and not per page, right?

If so, not sure this would work well, one document can have 1, while the other might contain 300...

[takolota]

@Hiwm If you want to avoid declaring 300 fields, you could use this template to OCR the document text, feed it to a GPT prompt, & ask the GPT prompt to return an array of customers / array of customer JSON objects.

https://powerusers.microsoft.com/t5/Power-Automate-Cookbook/Extract-Data-From-PDFs-and-Images-With-G...