Page MenuHomePhabricator

Allow list admins to train spam filters
Open, MediumPublic

Description

I am simply wondering the latter option is not available in Wikimedia's Mailman UI.

Screenshot from 2020-02-04 15-40-30.png (432×1 px, 101 KB)

I am aware of https://wikitech.wikimedia.org/wiki/Mailman#Fighting_spam_in_mailman which does not mention Bayesian classifications.
Some random further info seems to be e.g. https://spamassassin.apache.org/full/3.0.x/dist/doc/sa-learn.html or https://gist.github.com/benwebber/5631609

Event Timeline

@Aklapper we will have to dig into this a bit, thank you!

jijiki triaged this task as Medium priority.Feb 5 2020, 6:50 PM
jijiki added a project: serviceops.

Oh darrn! Thanks Reedy! I never realized that this is a custom patch in GNOME's Mailman instance, sorry!
Feel free to decline if this is too much maintenance costs.

Legoktm subscribed.

At this time we have no intent on carrying local mailman patches that haven't been merged upstream.

I found https://gitlab.com/mailman/mailman/-/issues/88 which is along the same lines as this request. Feel free to re-open this if you'd like to track that upstream request here too but I don't see much value since it doesn't exist at all yet.

Old mailman was able to forward spam to an email address (supposedly the admin), and I have been using it on my old lists to forward spam to my spam-learning email address. I do not see this in Postorius, but it would be a generic solution.

Even more generic solution would be to be able to move the spam into a maildir++ folder locally and then process these as the local ops would like, feeding back to sa-learn or rspamd. (I actually use this method, but mine is not a clean upstream-only solution.)