Fighting tracker spam with SpamBayes
During the last few days I've had time to do some programming just for fun. When you combine vacation and bad weather, you can be very productive :-).
Most of the time, I've been doing work for the roundup tracker instance for the python development team. I wrote about the new importer earlier, and now I've also created an anti-spam system based on SpamBayes.
For those interested, there's a technical description of the roundup spambayes integration in the roundup wiki.
SpamBayes seems to be a nice piece of software, especially now when it has an XMLRPC interface. Imagine having a SpamBayes XMLRPC server on your network, and then a plugin in your mail user agents that calls it to rate messages before sorting them down to folders, and a button in the user interface that allows users to report messages as being spam or legitimate content. That would be very powerful and give very few incorrect ratings as each organisation's SpamBayes server would learn what's legitimate content for the organization where it's installed.
When to sort?
As all users receiving large amounts of mail (in my case mostly because of mailing list subscriptions), I sort my mail. For my personal mail, I let the Cyrus IMAP server sort whenever the messages arrive, using the Sieve sorting language. I use the Sieve filter interface in Squirrelmail to create rules. This is very convenient as the mail is always sorted when I arrive at my mail client, regardless of which client I use. Also, I only have to define my filter rules at one place. I only wish there were more mail user agents with decent Sieve filter support.
- User A, who likes to arrive at the office early in the morning, opens his INBOX and find 2 messages that have been incorrectly classified as legitimate mail. He presses the 'report as spam button', which in an ideal world will teach the local SpamBayes server to score the message better.
- User B, who is a lazy bastard, arrives two hours later. When his mail user agent sorts his mail using the local SpamBayes server, he can benefit from the work made by User A earlier in the morning, as the two messages are now correctly sorted.