Document Processing text mixed up

[Fitchet86]

Hi all,

I've recently started using AI Builder at my company to automate part of the process where we receive PDFs into the business. Everything in it is working fine except one field where it is jumbling up the text. Below is how it shows in the PDF, but it has pulled it out as 43675395XY1.

On this one only 1 letter is in the wrong place, but on a another its mixed up more of it (Its meant to be X43738379Y1 but its pulled through as X433Y738791).

Has anyone seen this before, and know what is causing it?

Thanks!

[Antrod]

Hi @Fitchet86 ,

Do you get this issue when training the model or when doing a prediction? If it's during prediction, perhaps adding more samples in the training set could help.

If during training you already get the issue, it may be due to the boxes that may cause some issues. A solution could be to tag different fields for each box sequences and regroup them after prediction.