Best approach in Model selection/training?

Help Document Understanding
1

I am building a new automation for a client. When using the Extract Document Data activity in Studio, there are basically three paths that I can choose:

A) Use a Predifined extractor, for example Purchase Orders

B) Create a new Project in Document Understanding, and build a Custom model and train it from scratch, deploy it and then use that from within my activity.

C) Create a new Project in Document Understanding, and build on an existing model, again Purchase Orders as an example, do a bit of training on that, and then deploy the model and use from within my activity.

Over time, my model will need to extract Purchase Orders from many different companies, so many layouts, and also in different languages.

Which approach would you recommend?

Cheers,
Jeppe

2

@jjes

Option 3 would be more viable

That way you can get the leverage of already existing fields which reduces the annotation and any extra fields can be added if present

If you see no extra fields are need and if 1 is working for all types(which generally would not be the case) then go with 1

We are currently trying with 3rd option and i would say it is going good…and it is for purchase orders itself

Cheers

1 Like