Community filters

| 1 Comment
The admin of the student email system related a tale the other day that I found interesting. They use dspam for their anti-spam needs, and it has a Bayesian filter. It also has some other features which have, as I said, interesting side-effects.

There is a local independent movie theater that sends out a newsletter. Some students have plonked the newsletter into Spam rather than unsubscribe.

The dspam system is configured so that if enough students mark a specific sender as spam, then that sender is blacklisted system wide.

You can see where this is going? I thought so. Enough students have reported this independent movie theater's newsletter as spam that the whole system now blocks it, and we're getting reports of 'false positive!'

1 Comment

Funny you should mention this, Slashdot covered an article about legitimate email being marked spam yesterday.Using a system-wide, user-nominated blacklist doesn't seem like it would make much sense for a campus situation. Blacklist based on sender's IP? Botnets make that futile. Blacklist based on from: address? lol. Blacklist based on message content (eg throw the message into the spam corpus)? At first this doesn't seem so bad, but with so many spam messages using randomized english generation with a common vocabulary, it seems counter productive. Somehow correlating "spam-marked" messages by count seems like it would be useful, not many legitimate emails will come from a common source IP with identical content. I suppose they'd float to the top, and could be whitelisted.I'm glad there is finally good spam filtering, tweaking is easy compared to convincing the PowersThatBe to implement it.