Hello everyone,
we just got our instance of Decipher running since a few days and I am experimenting with the first document type. Glad that I found this thread, as I encounter the same issue Tejaskumar brought up.
I have a simple form providing bank account data. Every field is correctly caught by Decipher except for "Land:", as seen in below example.
I tried to train this document type dozens of times with slight variations but the error persists. The field "Land:" has the supposed value "Deutschland", but Decipher would always look down to the next header "Kontoverbindung" first, instead of using "Deutschland". I even tried to forbid "Kontoverbindung" by putting this word in the DFD in the "Exclude Values"-column - which seems to have no effect at all.
Which makes me even more mad: If I try Decipher on documents that have a different value like "DE" oder "Niedersachsen" then Decipher correctly identifies the value.
@BenLyons, in your last post you described a solution which refers to using different headers. I don't think that answer is related to what Tejaskumar and I are experiencing. We would need a solution that steers Decipher to prefere the direction "right first" instead of "bottom first". Is there any way to achieve this?
------------------------------
Til Minet
RPA Developer
EWE AG Germany
Oldenburg
------------------------------
we just got our instance of Decipher running since a few days and I am experimenting with the first document type. Glad that I found this thread, as I encounter the same issue Tejaskumar brought up.
I have a simple form providing bank account data. Every field is correctly caught by Decipher except for "Land:", as seen in below example.
I tried to train this document type dozens of times with slight variations but the error persists. The field "Land:" has the supposed value "Deutschland", but Decipher would always look down to the next header "Kontoverbindung" first, instead of using "Deutschland". I even tried to forbid "Kontoverbindung" by putting this word in the DFD in the "Exclude Values"-column - which seems to have no effect at all.
Which makes me even more mad: If I try Decipher on documents that have a different value like "DE" oder "Niedersachsen" then Decipher correctly identifies the value.
@BenLyons, in your last post you described a solution which refers to using different headers. I don't think that answer is related to what Tejaskumar and I are experiencing. We would need a solution that steers Decipher to prefere the direction "right first" instead of "bottom first". Is there any way to achieve this?
------------------------------
Til Minet
RPA Developer
EWE AG Germany
Oldenburg
------------------------------
Hi Til,
There's no way to specifically tell what direction for Decipher to look in, but training of the particular document layout should effectively train the correct location.
However, other things you can try in this situation is adding "Kontoverbindung" to excluded values for the field. This should get Decipher to ignore it, though it may not result in Decipher picking the correct value.
If the potential values for "land" are limited in any way, you could provide values to the options parameter e.g.
You shouldn't need to select any additional flags, unless you want the field to appear as a dropdown box (selectable).
Let me know how you get on.
Thanks
------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------
There's no way to specifically tell what direction for Decipher to look in, but training of the particular document layout should effectively train the correct location.
However, other things you can try in this situation is adding "Kontoverbindung" to excluded values for the field. This should get Decipher to ignore it, though it may not result in Decipher picking the correct value.
If the potential values for "land" are limited in any way, you could provide values to the options parameter e.g.
You shouldn't need to select any additional flags, unless you want the field to appear as a dropdown box (selectable).
Let me know how you get on.
Thanks
------------------------------
Ben Lyons
Senior Product Specialist - Decipher
Blue Prism
UK based
------------------------------
Hi Ben,
Thanks for your reply!
After 4 more batches with around 15 documents each I get the feeling, that Decipher has a specific problem with the word "Deutschland". As soon as the field "Land:" has a value that is not "Deutschland" I get a match. If the value consists of "Deutschland" and some other text Decipher would only return part of the field and exclude "Deutschland".
The error with a match for "Kontoverbindung", which is a different header, persists even though I excluded the value "Kontoverbindung" in the DFD.
In my case it would be an option to make "Deutschland" the default value or even use a list of possible values. But nonetheless - I find this behaviour very interesting and am worried that I encounter such behaviour again in a different DFD where I can not use a workaround.
Look at below examples:
First of all - that's how I excluded "Kontoverbindung" from the values. Sidenote: "Deutschland" is not excluded.
That's what's Decipher thinks of my intention. The capture stage was definitely done after the DFD config was saved. This is reproducable.
This happens regularly... I've got plenty of those.
Land is written as "Country / state" - only the state is matched. Even though the region includes "Deutschland".
Different format, same behaviour. No Germans allowed in here. 😉
Land is "DE" - perfect match
Land is "Deutschland" - Sometimes it's just no match at all. If I point at the right region Decipher has no problem to identify the text, though it's written rather small.
State as given value - match.
Handwritten form - Decipher would hit the right region
Can you explain that behaviour?
Best regards
Til
------------------------------
Til Minet
RPA Developer
EWE AG Germany
Oldenburg
------------------------------
Thanks for your reply!
After 4 more batches with around 15 documents each I get the feeling, that Decipher has a specific problem with the word "Deutschland". As soon as the field "Land:" has a value that is not "Deutschland" I get a match. If the value consists of "Deutschland" and some other text Decipher would only return part of the field and exclude "Deutschland".
The error with a match for "Kontoverbindung", which is a different header, persists even though I excluded the value "Kontoverbindung" in the DFD.
In my case it would be an option to make "Deutschland" the default value or even use a list of possible values. But nonetheless - I find this behaviour very interesting and am worried that I encounter such behaviour again in a different DFD where I can not use a workaround.
Look at below examples:
First of all - that's how I excluded "Kontoverbindung" from the values. Sidenote: "Deutschland" is not excluded.
That's what's Decipher thinks of my intention. The capture stage was definitely done after the DFD config was saved. This is reproducable.
This happens regularly... I've got plenty of those.
Land is written as "Country / state" - only the state is matched. Even though the region includes "Deutschland".
Different format, same behaviour. No Germans allowed in here. 😉
Land is "DE" - perfect match
Land is "Deutschland" - Sometimes it's just no match at all. If I point at the right region Decipher has no problem to identify the text, though it's written rather small.
State as given value - match.
Handwritten form - Decipher would hit the right region
Can you explain that behaviour?
Best regards
Til
------------------------------
Til Minet
RPA Developer
EWE AG Germany
Oldenburg
------------------------------
