Hi Tejaskumar,
I can't believe I missed your original post! My bad.
Great question and I can certainly give you some info on how to figure this out. This will largely only be relevant to structured documents, i.e. any document where the data appears in a consistent/semi-consistent layout.
So the rules based ML engine is always on, you can't disable it and makes up the Training Data. This can be pretty powerful by itself and for structured documents may get you the results you need. This works by utilising the 'hints' found in the DFD such as keywords (e.g. sample headers), data types, lists, regex and formulas. Following verification of your first document, it begins enhancing these hints with location data.
All of this is stored based on layout, so the training/learning is directly linked to what the document looks like. Each time Decipher receives a document, it looks through the Training Data to see if it recognises it. This is great because it means what you train with Layout A won't impact training for Layout B and so on. This is not linked to the DFD, so it can be shared between processes if needed without additional configuration.
So, if you're getting your target results, great, you can leave it 'as is'. But if you need to stretch your results a bit higher, you can start to train your Capture ML model. This will be specific to the Document Type/DFD, and will create a more sophisticated model which extends beyond the capabilities of the native training data. This is only designed to improve upon a well trained data set and will not help in the event of low initial accuracy.
So if you're looking to take your success from 85%, further, you can enable this feature and you may see your results go to 90%+. It helps by improving the extensibility of the training, so the more documents you train the better it will get. So it's a good idea to start with a higher training count, this will help ensure the model is based on a wide selection of documents.
We've just updated our
best practice advice, which will provide a bit more detail on how you can adapt your approach.
Thanks
------------------------------
Ben Lyons
Product Consultant
Blue Prism
UK
------------------------------
Ben Lyons
Principal Product Specialist - Decipher
SS&C Blue Prism
UK based
