Anti-spam algorithms & MIT conference papers.
CRM114 - the Controllable Regex Mutilator is a spam filter that claims 99.9% accuracy.
DSPAM is another scalable, open-source statistical-algorithmic hybrid anti-spam filter.
Source code is available for both. Both are meant to integrate with unix flavor mail servers. One wonders whether the spam / anti-spam war isn’t much like the virus / immune system: Spam will adapt to each new filtering algorithm; all victories will be temporary.
The end result, of course, will be spam that passes the Turing test and anti-spam filters that can measure intelligence. Which is harder?