Perfecting Spam Filtering Using Game Theory
from the costs-and-benefits dept
The trick with spam filters is always how fine do you tune them against Type I and Type II errors? Would people prefer that the spam filter be more aggressive and block the occasional legit email in favor of keeping more spam out of your inbox, or do they want to make sure that the legit emails all get through, even if it means a few extra spam messages get through as well? Apparently, some researchers are now trying to model that question using game theory — putting a price on each of the different kind of errors. However, they also are modeling in the cost and benefit to the spammer. So spam that gets through your filter is a cost to the user, but a benefit to the spammer. They’re hoping that this model will help ISPs and anti-spam operations better tune their filters to provide greater benefit to end users and higher costs for spammers. It’s a good idea — but the problem is that it’s not clear that the costs and benefits are the same across different users (or emails!). Certain legitimate emails aren’t a huge deal if they’re caught by a spam filter, but others can be quite a nuisance.
Comments on “Perfecting Spam Filtering Using Game Theory”
No Subject Given
I hope the paper itself is a bunch more useful than those summaries of it. I’m well aware of the game I play with choosing to report various mails to spamassassin or not, to the point I’d say that it looks like typical New Scientist gloss, until I see the predictions from a useful theory.