[2023-02-01 02:05:00,186] [DEBUG] [paperless.classifier] Document classification model does not exist (yet), not performing automatic matching.
[2023-02-01 02:05:00,188] [DEBUG] [paperless.classifier] Gathering data from database...
[2023-02-01 02:05:03,604] [WARNING] [paperless.tasks] Classifier error:
  Resource stopwords not found.
  Please use the NLTK Downloader to obtain the resource:
       import nltk
       nltk.download('stopwords')
  For more information see: https://www.nltk.org/data.html
  Attempted to load corpora/stopwords
  Searched in:
    - '/usr/share/nltk_data'
The same appeared when I invoked id manually via manage.py:
  root@server:/opt/paperless/src# sudo -Hu paperless python3 manage.py document_create_classifier
[2023-02-01 01:55:03,211] [WARNING] [paperless.tasks] Classifier error:
      Resource stopwords not found.
      Please use the NLTK Downloader to obtain the resource:
          import nltk
          nltk.download('stopwords')
  For more information see: https://www.nltk.org/data.html
  Attempted to load corpora/stopwords
  Searched in:
    - '/usr/share/nltk_data'
I don't know what went wrong when I installed it. Maybe someone can review the installtion instructions and check.
However, i just wanted to share the workaround I used:
Create the directory:

sudo mkdir /usr/share/nltk_data

sudo chwown paperless:paperless /usr/share/nltk_data
Download the required data (It turned out that I also needed a ressource called "punkt", might be because I have documents in german. You will see what you need from the error messages.):

sudo -Hu paperless python3 -m nltk.downloader  stopwords

sudo -Hu paperless python3 -m nltk.downloader  punkt
Renew the classifiers:

cd /opt/paperless/src

sudo -Hu paperless python3 manage.py document_create_classifier
That woked for me, mybe it's useful for someone else.
        
(I'm posting links here for other folks to reference.)
I ran into this same problem.  The docs really do need to be improved in this regard.  It's mentioned in step 15 of the bare metal docs, but the CSS makes it really difficult to tell the hyperlinks from the text.  Linking to the nltk docs is.. nice, I guess, but it's not obvious what to do.  Explicit is better than implicit.