Can SpamAssassin really be trained?

backpackhasjetz 2017-05-02 17:02:49 UTC #1

There's no documentation that I can find for Mail-in-a-Box that defines this, but numerous folks have said that simply moving an email to the "Spam" folder should have SA auto-learn it as spam. And possibly moving it from Spam to Inbox does the opposite (learns it as good/ham?)

But I want to know is that really true? I've sent so many emails to the spam folder, it doesn't seem to ever improve. I'll get the same spammy "check out my videos" messages right to my inbox.

elser 2017-05-02 18:31:58 UTC #2

Its setup here.

github.com

mail-in-a-box/mailinabox/blob/master/setup/spamassassin.sh#L69


# but it's also annoying to get to the original message when it is an
# attachment, modern mail clients are safer now and don't load remote
# content or execute scripts, and it is probably confusing to most users.
#
# Tell Spamassassin not to modify the original message except for adding
# the X-Spam-Status mail header and related headers.
tools/editconf.py /etc/spamassassin/local.cf -s \
	report_safe=0 \
	add_header="all Report _REPORT_"


# Bayesean learning
# -----------------
#
# Spamassassin can learn from mail marked as spam or ham, but it needs to be
# configured. We'll store the learning data in our storage area.
#
# These files must be:
#
# * Writable by sa-learn-pipe script below, which run as the 'mail' user, for manual tagging of mail as spam/ham.
# * Readable by the spampd process ('spampd' user) during mail filtering.
# * Writable by the debian-spamd user, which runs /etc/cron.daily/spamassassin.

JoshData 2017-05-02 21:13:31 UTC #3

It's not clear that this is working at all. On the other hand, it's not clear that it's not working either. I don't know that learning is working on my box, but spam filtering on my box is extremely good.

michaelkroes 2017-05-03 08:43:31 UTC #4

I think it is working. When I send a message to spam, similar message are later automatically send to spamm. The bayesean database is also updated. But proving it is another thing.

CorneliusLentulus 2017-05-03 10:45:16 UTC #5

I have the same experiences @michaelkroes. There are some plugins around for Roundcube on GitHub (although most still in beta) that have SpamAssassin implementation learning possibilities (sa-learn --ham --username=%u %f or sa-learn --ham --prefs-file=/var/mail/%d/%l/.spamassassin/userprefs %f_ ).

wuxmedia 2017-05-13 08:07:09 UTC #6

Haven't had any spam yet as all new addresses, I got spam learning to work on my previous box, as previous poster said need to get it to run sa-learn with the 'spam' folder - the it learns and marks the spam, the tricky bit was getting sieve to move it into the 'spam' automatically,
So I had spam all marked up nicely, but loose in the inbox, some users made a filter in their mail client.
PS the ham is not spam,

Richard5 2017-05-18 07:18:51 UTC #7

You can always check the headers in each email to see the workings and conclusions that Spamassassin draws from your email history. Look at the "X-Spam-*" header in each mail, those might indicate why mail is indicated as ham or spam.