Everyone's talking about auth to limit spam, but I think a more interesting approach is to let people click to flag spam, then train up a little naive Bayes classifier.