Section on : Counter-Attack

1 : Spamalot: A Toolkit for Consuming Spammers’ Resources; Peter Nelson, Kenneth Dallmeyer, Lucasz Szybalski, Tom Palarz, Michael Wieher – A very fun presentation regarding using an AI system to use up and abuse the 419ers and loan spammers time with a 65% success rate and a average chain length of email 10 mail exchanges. Go read the paper. Though they know what happened to Blue Security.

2 Barry Leiba from IBM research – Breaking Anti-Spam systems with parasitic spam (p-spam) , It’s also in todays New Scientist. A new (? Yahoo have done this for years ) model for spam. Attaching the spam to the bottom of a legit message and how products would deal with it. This is not the first time I’ve heard the idea bounded about. Infecting an company or ISP MTA and rather than spewing wildly this technique adds collateral damage to their spam. The presentation includes suggestions for counter measures.

3 Steve Webb on Observed Trends in Spam Construction Techniques – So it’s an arms race… They studied features that stopped and continued on regardless of detection. The techniques that expire over time appear to be the simple to changes ones, rather than the features inserted by the spammers tools. They do note that some techniques die not because of spam detection changes but sometimes external changes such as browser releases. Nice interesting research, however it’s suffering from the time machine effect of scanning 3 year old mail with a new version of SA.

Section on : Learning-based Filters I

4 Spam Filtering with Naive Bayes – Which Naive Bayes?; Vangelis Metsis, Ion Androutsopoulos, Georgios Paliouras. Ion is presenting a comparison of bayes methods/tools. This is a high end study on the application of bayes approaches, a must read for anyone that uses bayes. Very interesting.

5 An Adaptive, Semi-Structured Language Model Approach to Spam Filtering on a New Corpus; Ben Medlock, University of Cambridge. Ben gave a good presentation on modeling corpus for training, adaptation and testing purposes and their ILM classifier.