Hi, after using Docspell for a while now I would have 2 feature ideas regarding the machine learning capabilities.
- it would be nice to have support for custom fields as well. For example, if you label a lot of your documents with an ASN, it would be cool if that was also automatically entered into a custom field.
_(info about ASN)
https://paperless-ng.readthedocs.io/en/latest/usage_overview.html#processing-of-the-physical-documents
For each document, decide if you need to keep the document in physical form. This applies to certain important documents, such as contracts and certificates._If you need to keep the document, write a running number on the document before scanning, starting at one and counting upwards. This is the archive serial number, or ASN in short.
Scan the document.
If the document has an ASN assigned, store it in a single binder, sorted by ASN. Don’t order this binder in any other way.
If the document has no ASN, throw it away. Yay!
2 replies
- you could add another field for the date of birth when creating a person. From time to time, the date suggestion also shows the date of birth of a person, as a suggestion. If the date of birth is stored with the person, it could be filtered out in the date suggestions, when adding a document.
However, since it does not actually add any extra work when sorting documents, this feature is rather less important.
2 replies
- Lucki Yeah, doesn't have to be, but I kind of thought it would be a useful value to add.
( You could also say that you search for the company name with the full text search every time to find a document :) )
eikek Exactly invoice amounts would be a similar use case. But probably a bit more complicated. I had imagined that it could be possible for machine-learning function to recognize the pattern of a 5-digit number, which is usually at the top of the page (top left). And the regularity that this is always entered in a certain field :)
lukas2: Yes, detecting the invoice amount is indeed much more complicated, and you are right that it is probably relatively simple to recognize the asn number even when only looking at text. It would require to configure which field to set, though, and so I think I'll add it later as a separate plugin maybe; but I'm not sure at all currently - only the current state :).
Regarding dates: Yes, these dates should really be sorted in the suggestions! The are sorted for the due dates, but not for the item date… good catch!
Another suggestion for the future :D
It is currently too much work for me to assign a meaningful title to each document. If the position on the page could be used, then it could be possible to make a suggestion for a title (if no meaningful title was assigned)
Since the title line is often in the same place on a letter and may also be in bold print. Maybe there are already other tools that could add such a feature in the future. However, I have never searched to see if that might exist.
1 reply
If the documents are filtered using one of the meta fields (tag, organization, etc.), all other fields should only contain pre-filtered entries in order to only receive possible hits.
I have a lot of correspondents with spaces. When I search for an entry in the edit mode, after entering the space, the search field is jumped and I have to go back in to be able to continue writing. That is a bit exhausting in the long run.
I don't quite understand the results. Shouldn't "A_Org" come earlier in this case and shouldn't be displayed as the only result after entering the entire string?
All right. Something else that caught my eye and that annoys me a bit;).
I have a lot of correspondents with spaces. When I search for an entry in the edit mode, after entering the space, the search field is jumped and I have to go back in to be able to continue writing. That is a bit exhausting in the long run.
The problem does not seem to exist on the smartphone as I have just noticed