Colin - 2007-05-25


Just a question. Can popfile be configured to run several classifiers on an email, one after another?

I've found that when deciding between several "buckets" the accuracy of bayes systems isn't so good, even after loads of training. So I thought it would be better to serialise the classification process.

Instead of asking

Is this mail Junk, Personal, About_Karate, About_Business or From_Wife?

You ask

Is the mail Junk? yes/no
Is the mail Personal yes/no
Is this mail About_Karate yes/no
Is this mail About_Business yes/no
Is the mail From_Wife yes/no

etc etc with an arbitrary number of classifiers being run in series on every incoming mail. The accuracy of this would be far far better than the regular method. The mails could then be tagged with all of the classifications that they had been given. Adding/modifying the relevant tags at the web interface would update the relevant corpus.