If You're Going to Read One Academic Anti-Spam Paper this Year ....

By George Bilbrey
President

I have a difficult admission to make: I read a fair number of academic anti-spam papers.  We are constantly on the hunt for ideas that can make the reputation systems that Return Path runs a little bit better.  There are a lot of people doing some really clever stuff out there.  There are a lot of people who are sure that they have the "Final, Ultimate Solution to the Spam Problem" (FUSSP) which only requires everyone to change how they handle mail.   Rarely, however, have I been as impressed with an anti-spam paper as "Spamlytics: An Empirical Analysis of Spam Marketing Conversion" by a group of researchers from the University of California, Berkeley and the University of California, San Diego.  More than anything, I love the audacity of their project.   

A brief summary of the project:

* The research team actually "hijacked" part of the Storm Botnet
* They modified the outbound spam from their part of the botnet to contain safe links that went to safe sites
* They measured delivery rates and conversion rates across three "campaigns" (offers) that were being sent out through the botnet.

The results are pretty interesting,  in summary:

• Response rates are very, very low.  So low in fact that it makes you think about how much money there is in spam.  For hundreds of thousands of messages sent, there are tens and hundreds of "conversions."    When I do my own back of the envelope math, there has to be a ridiculously low cost for renting botnets to make the economics go round.  The authors conclude that one way that the economics could go round is for the operators of the botnet to actually "own" the offers.
• Very little of this mail gets through.   When they point some of these messages to seed accounts, almost none of it gets through with really good performance by Hotmail, Yahoo, Gmail and Barracuda.

So, if you're going to read one academic anti-spam paper this year, I'd recommend this one.