Errors in Data Verification and extraction

[Venkata_Pranav_]

We have a DFD mapping say 100 fields for a type of document. Some of the documents we need to extract from have only 80 fields. We get errors every time when these kind of documents sent into Decipher. 

How much training are we expected to perform? We tried several things, enabling machine learning, retraining the documents, but to no effect. We still get random hallucinations where Decipher randomly maps a different field. 

Is anyone else experiencing this? 

------------------------------
Pranav
------------------------------

[BenLyons]

Hi Venkata,

They're probably not "hallucinations", but Decipher using the DFD configuration/training you've set up to find the desired field. So if you've got a Sample Header/Keyword specified for a field, Decipher will automatically use these to identify the field, even on previously untrained document layouts. If a keyword is used in a document, but the respective value isn't there, Decipher can be trained that it doesn't exist in most cases. If the keyword is used differently in some document, it might be better to remove it and allow Decipher to be trained using its training data machine learning algorithm.

Alternatively if it's happening in a trained layout, and it's a case that sometimes the field is there, sometimes it's not, either it needs more training or perhaps some changes to the validation. After 10 or so document s of the same layout/from the same source, you should be getting pretty consistent results. Could you add anything to the Format expression field? Maybe a list?

When you've tried different configurations have you been deleting your old training data? This would still influence the outcome even after DFD updates and further training.

Thanks

------------------------------
Ben Lyons
Senior Product Specialist - Decipher
SS&C Blue Prism
UK based
------------------------------