Flaming: A White Paper
I was reading a worst-case scenario guide on how to expunge a nasty e-mail and got to the part where it mentions automatic flame detection:
One e-mail program offers a “Mood Watch” function that monitors your typing and alerts you if a message is approaching “flame” status.
I vaguely recalled this as a feature of an e-mail client I used, so I did some Googling and found that MoodWatch is a feature of the now open source Eudora client, which I used until switching to Mac OS X.
As it turns out, they’ve got a white paper on the algorithm. Essentially, they dug through alt.flame and categorized “words and phrases that are commonly considered offensive, dictatorial, aggressive, insulting and rude.” The authors come just short of proposing a Bayesian spam flame filter, but I imagine that’s how it’s implemented in Eudora (I didn’t trudge through the source).
It gets me thinking, though, why we couldn’t use Bayes’ theorem for sorting all kinds of e-mail. We don’t have to use a black-and-white differentiation between spam and non-spam. To be sure, 419 frauds, phishing e-mails, and unsolicited stock advice is spam through and through. But that Amazon.com sale might be useful to me, I just don’t want it in my inbox.