Dear community members,
we try to run Decipher PoC in these days.
We have no experience with OCR tools. In our PoC, we have a document type that contains identical data, but is very variable in its structure. We assumed that the machine learning built-in Decipher should cover this varibility.
We are not happy with the results of PoC. Perhaps it would help to better understand how machine learning works in fact.
What I doesn't understand is:
- In the Decipher settings I can define the number of documents for creating the machine learning model. Until this limit is reached is ML model not active? In other words so far we see, when verifying data, only the results of rule-based assignment of region fields to in the document form definition defined fields ?
- Is ML only learned from user corrections of extracted data in the validation form? Do it also learn from resizing highlighted regions on the document when they are not focused accurately or correctly?
- Highlighted regions are determined by how the structure of the document understands OCR and ML does not interfere in this process?
Thank you for your answers.
JH
------------------------------
Jiri Hlucil
Blue Prism Developer
Sberbank CZ, a. s.
Europe/Prague
------------------------------
