El 1/2/21 a les 10:48, Ian Batten via Info ha escrit:
Is it worth doing at all?
I run rspamd outside Cyrus: it’s a Postfix milter, Cyrus gets an lmtp
feed but has no interaction with rspamd.
Well, it depends. I'm still using spamassassin (I'd like to try rspamd
but I didn't yet) and bayesian filtering is just a small part of the
scoring, so maybe it doesn't make much difference.
At home I have a simpler setup, I use dspam (sadly it's discontinued)
and it's really effective, but it needs training.
I’m also nervous about using the user’s spam folders because those are
often the result of client-side Bayesian filtering, which is somewhat
error-prone, so the chances of ham (or at least bacn) ending up in there
is non-zero. Users are also very bad at classifying spam so using their
main inbox as a source of ham is unwise.
That could be a problem with spamassassin (the training is server level,
but then the remaining rules have more weight than the bayesian
filtering) but not with dspam (each user has its own dataset).
Rspamd is _/very/_ good, and increasingly I just leave it to do its
thing. With rspamd and clamav’s phishing patterns I doubt I see a piece
of phishing email getting through to me more than once a month, and my
users report similarly. I have a lot of bacn which is correctly marks
as spam-ish, and I similarly see very little outright spam.
Before wading in an doing a complex learn/mode thing, just run rspamd
unsupervised for a month and see how it does on its own. I think you’ll
be very pleased.
I'll have to try it one of this days.
Bye
--
Luca Olivetti
Wetron Automation Technology http://www.wetron.es/
Tel. +34 93 5883004 (Ext.3010) Fax +34 93 5883007
------------------------------------------
Cyrus: Info
Permalink: https://cyrus.topicbox.com/groups/info/T4a9f972d95fd358f-M9c66bad1450dc862857aa607
Delivery options: https://cyrus.topicbox.com/groups/info/subscription
