Email Classification from Good to Bad

If you've never read Paul Graham's classic A Plan For Spam, you may find it interesting. In a nutshell, filtering spam amounts to estimating probability of spam based on the proportions of "spam-like" words and "non-spam-like" words in a message. Messages receive a score and based on that score, a line is drawn and a call is made: spam or not spam.

Rather than a binary yes/no, more and more I find myself wishing I could score my email on a scale of desirability / importance. A scale from -10 to +10 might work. Negative scores would indicate spam and such as being completely undesirable and unimportant (toss it!). High positive scores would indicate something that's really important. As with spam filters, with training, the system would learn to score messages automatically.

Having the classifier support more classes (-10 to +10 or whatever) would make it possible to, for example, set a threshold for those annoying little Outlook pop up messages that occur in the lower right corner, which I don't shut off completely because they offer the benefit of alerting me to important messages at the cost of alerting me to the unimportant messages in the majority.

Desktop interruptions seem to be a growing problem. What's really needed is the equivalent of a virtual receptionist that can intelligently screen interruptions and prioritize incoming information.