I'm playing around with spamd for a few accounts on our mail server via procmail and I would like to set up bayesian filtering. Everywhere I read says that per-user is the way to go with bayesian filters versus one big site-wide database, because everyone gets different spam and ham, etc., though I seem to get the same spam as everyone else I know.
Well, the impact is the kind of ham you get. Pretty much _everyone_ gets the same spam.
And really, a single-database site-wide bayes isn't all that bad, particularly if your users get similar types of normal mail. The biggest factor is getting nonspam email that represents all your users. And the drawback is that the broader your range of nonspam mail, the greater a chance there is for spam to avoid the bayes database.
In a small-company type environment, this is really quite practical. Most of your users are going to get similar mail pertaining to your area of business, and some run-of-the-mill type personal email. It's not quite as practical as a large ISP, because the breadth of nonspam email is pretty wide, but even there it's not so bad.
So I wouldn't chalk up the single database idea as not workable, but you certainly will get better accuracy in many situations out of per-user training.
Is there any way to train the filter on a per-user basis without handing out login shells?
More importantly, is there a way to even have per user databases without handing out login shells... This can be pretty tough to pull off...
My initial suggestion would be:
1) use per-user user_prefs from mysql, and have those user_prefs specify alternate bayes_paths on a per-user basis
2) use the --dbpath option to sa-learn to force where the database that is learned to is located.
so you could then have only different bayes directories per-user, instead of shells, like these:
/var/bayes/joe/
/var/bayes/jendries/
/var/bayes/mkettler/
------------------------------------------------------- This SF.net email is sponsored by: The SF.net Donation Program. Do you like what SourceForge.net is doing for the Open Source Community? Make a contribution, and help us add new features and functionality. Click here: http://sourceforge.net/donate/ _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
