Ricardo Kleemann wrote: > I'm running spamc/spamd 3.2.4 on a Ubuntu 8.04 server, it's the > standard Ubuntu package. I have the default settings for Bayes (with > auto_learn) and I'm using a mysql backend for BayesStore.
It’s worth noting that Bayes, by itself, is not allowed to condemn spam. Its maximum score, by default, is 3.5. That means that if you’ve got the standard required_score of 5, the spam will not be marked unless it also hits other rules with a combined score of at least 1.5 – even with perfect training. If you are confident in your training (all spam gets learnt as spam, and all non-spam gets learnt as non-spam) you may wish to try raising the BAYES_99 score. Note that, in theory, 1% of email hit by this rule is non-spam – but also that another theory says that the maths behind that figure are wrong. It can be difficult to stop “narrowcast” spammers that spam a limited mailing list (perhaps industry specific), using standard bulk-mailing tools, and who don’t get onto the various DNSBLs. Their spam looks like solicited bulk email, there’s practically nothing about the spam that distinguishes it from wanted email that isn’t specific to that particular spammer, and they stay out of the notice of antispam organisations. If you want to stop spammers like these, you may have to create specific rules (either sending domains, or sending IP address, or specific phrases in their emails – or best of all, all three, with each rule attracting a few points). The good news is there aren’t that many of these spammers, and they tend to try to look like a legitimate company, so they will usually send using a constant identity, probably use a constant domain, and they’ll either use their own servers or e-mail service providers. So static rules aimed at them can be very effective. Alternatively, you could post sightings on news.admin.net-abuse.sightings and see if Spamhaus pick up on them. That way, more people can be spared their spam. James. -- E-mail: james@ | Remember, half-measures can be very effective if all you aprilcottage.co.uk | deal with are half-wits.
