Pre - Publication Draft 21

MailCat is an intelligent assistant that helps users organize their e-mail into folders. MailCat uses a text classiier to predict where each new message is likely to be led by the user and provides shortcut buttons to quickly le messages into one of its predicted folders. MailCat operates in a dynamic environment in which it must adapt to the user's continually changing mail-ling habits. MailCat uses incre-mental learning to cope with these changes. This paper evaluates the performance of MailCat's incremental learning algorithm. We investigate how incremental learning performs over time during simulated usage. We nd that MailCat's incremental learning algorithm performs well. MailCat can almost instantly learn about a new user's mail-ling habits and maintains over an 80% to 90% accuracy level throughout our trials. We observe several interesting dynamical behaviors including an inverted learning curve and the ability to learn from single examples.