Mining Decipher Training Data For Fun and Profit

[stepher]

Salutations, All,

Preface:  I understand that the Training Data file that is downloaded from Decipher is a renamed Windows Compressed File folder.  Within this folder are the myriad definitions used to extract the data from the scanned documents.

Question: If I, or someone I knew, were sufficiently adept at querying the Decipher SQL database, is it possible to associate the specific Training Data Definition used to parse a specific document?

Use Case/Justification 1 - Surgical Excision of Errant Training Data:  I have a number of scenarios where Decipher is not quite getting what I want from the document.  Usually, this entails Decipher's boundaries being too large and grabbing more than the pertinent information.  It can also be that Decipher is just picking up the wrong values.  Try as I might, I cannot seem to correct this.  My thought is that I could extract the Training Data, delete the errant entries and then reload the refreshed Training Data.  When the documents in question come through the next time, I could mold the new training data in my image.

Use Case/Justification 2 - Removing Stale or Unused Training Data:  If some of the Training Data Definitions had not been used in a period of time, say six months, it seems like it could just be removed.

Of the two, the first case is my current priority.  I don't know that I would really get around to the second case.  (Wait, I am self-aware enough to know that my OCD would probably get the best of me.)

I think that about covers it.

Thanks,

Red 

[BenLyons]

Hi Red,

There might be some new features in Decipher 2.3 that could save you the trouble. We hope to release it by the end of next week, we are in the very final stages.

Roadmap disclaimer, this may not appear in the final release, but you can filter your training data in the UI by Created Date, Document type, Batch Type, Unique Field Value and Created By. You can then delete/export using this filter.

Feel free to reach out via your account manager and we can set up a demo in due course.

Thanks

Ben