In reality, I've observed that you can filter present-day spam acceptably well making use of almost nothing more than a Bayesian blend of the spam possibilities of personal words. Using a slightly tweaked (as explained down below) Bayesian filter, we now skip much less than 5 per 1000 spams, with phony positives. Merely seeking for the word "click on" will catch 79.7% of the emails in my spam corpus, with only 1.2% wrong positives.