Comment Where is the paper? (Score 1) 353
I can't find the paper anywhere. This is just hype!
Templates are relatively similar, however, the trick is in figuring out what the polymorphic engine that injects the content does.
Currently, there's quite a bit of research being done uses machine learning techniques. I guess this paper is supposed to do the same thing, learn from some training data and filter other spam.
So, there's a learning phase and a testing phase. Current research shows that for the short term these techniques are quite successful having really low false positives 0.002 (AutoRE at Microsoft). However, when looking at spam say 6 months down the road unless you keep learning from labeled emails spam/not-spam, you won't be able to make good decisions because of the volatile characteristics of spam.
Also, you've got to be concerned with the Real-Time implications of these learning methodologies and noise. Most campaigns last around 5 days, however, will a user have to wait for 5 days to receive an email? Probably not, so this technique may yield low positives in a longer period than the short term.