I find it amusing that they propose a Bayesian approach since PG contributed a fair deal to spam detection a decade or so ago.