Bayesian Spam Filters

Follow-Up:

Several folks wrote to warn me about the dangers of recommending “bouncing” spam back to the senders. To quote Roger Matus, whose company makes InBoxer anti-spam software:

Let me get his second point off the table, just because I disagree with it. Sorry! I just don’t believe that we’re talking about so much bandwidth as to make a difference. (There used to be an organization called the “Bandwidth Preservation Society” that offered suggestions like “do all your downloading late at night when demand is low” and other, similiar ideas. Silly then, silly now.)

But his first point has merit. As the owner of several domains, every now and then I see bounces coming to them — these are bounces caused by some spammer spoofing one of my domains. In other words, the junk looks like it’s coming from, say, whizkid.com, but isn’t. So I get the bounces. That seems to be a good argument against bouncing spam; half the time or more you’ll be annoying some innocent person.

But there is some merit to bouncing, just as there is some merit to that CAN-SPAM Act.

If you look through your spam (yes, that’s like the vet telling you to “examine the animal’s feces) you might notice there are two kinds. Type 1 spam comes from legitimate businesses that have deluded themselves into thinking they have a right to pitch you — maybe they’re the “business partners” of some store you shopped at. These are easy to spot: The subject line is plain as day, e.g., “Get your next car at the price you want!” or “Cheap cigarettes!” There might even be a working “unsubscribe” button and/or a valid postal address in the message.

Type 2 spam is the true, vile junk: Porn, generic drugs, and that ilk. These are the spammers I mentioned in my column that make their money just by your opening the message. These are the folks that hide their real mailing addresses, and that modify the subject line with random characters and other such things to try to fool spam filters.

There is no point in trying to bounce Type 2 spam (at least until someone comes up with a way of tracing the actual sender). It’ll just go to some unsuspecting schlub who’s domain is being spoofed.

But there is hope in bouncing that Type 1 spam. There’s a decent chance, as many of these are legit businesses, that your bounce will go to the right place: a machine that reads it and takes your address off the list, thinking that it’s no good.

So while the people who wrote to say that spam filters that bounce might not be a good idea are partially right, I think there’s a good argument for carefully bouncing your spam.