Manually writing values in Data Verification

[AlexandraBarban]

Hello, 

We have some problems with checkboxes in Decipher because it does not always identify them correctly (True/False checkboxes) when the annotators are selecting the region of the checkbox. 

My question would be if the annotators will select the right region, but will write manually the value (true/false), without drawing the rectangles until Decipher will correctly set the value, will this impact the training data? Will Decipher learn from manually written values by the annotators if the correct region is selected? 

Thanks, 
Alexandra Barbanta

------------------------------
Alexandra Barbanta
------------------------------

[Tejaskumar_Darji]

Hello Alexandra,

As per my testing with simple checkbox Decipher was able to figure out true and false properly based on the checkbox image.

Rather then forcing to update manually I would recommend re-drawing the checkbox region again and see if it grabs value automatically.

It might be due to the region drawing is incorrect at first place.

Check out this video where how to draw region is shown.

------------------------------
If I was of assistance, please vote for it to be the "Best Answer".

Thanks & Regards,
Tejaskumar Darji
Sr. Consultant-Technical Lead
------------------------------

[NileshJadhavBP]

Hi @AlexandraBarban 

this is first time i heard , can you please re-drawing the checkbox region again and check.

If still issue is there please raise ticket with Blue prism ( Product team will help you)

Thanks
Nilesh

------------------------------
Nilesh Jadhav
Senior RPA Specialist
ADP
India
------------------------------

[AlexandraBarban]

Hello, the region is correctly placed. There have been made lots of retries but because the documents are not having such a good quality, most are scanned, not 100% digitals, there are problems with these checkboxes. 

The question would be just the following: if the region is correctly surrounded by the rectangles but the value identified is for example "true", not the right one which is "false"  and the annotator by hand will just write "false", without redrawing the rectangles until Decipher set it to "false", will this make Decipher learn anyway? Or this will mess the already existing training data?

------------------------------
Alexandra Barbanta
------------------------------

[BenLyons · Accepted solution]

Hi Alexandra,

Decipher won't learn from manual changes to checkboxes. If it correctly finds the region it will assess whether the box is sufficiently completed or not. This can be an issue if the box has marks from a poor quality scan or similar.

The threshold can be adjusted using a misc parameter "RFT" (Region Fullness Threshold). You can then set it higher or lower as needed e.g. RFT = 5, I believe the default is 7%.

I think the percentage is displayed when you hover over it after selecting the regions, though I think that's new for 2.2.

Thanks

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