Experiments in multi-class e-mail categorization

This paper reports on extensive experiments in multi-class document categorization with supervised and unsupervised machine learning techniques. in particular, experiments on a document collection consisting of personal e-mail messages are described. Two distinct document representation formalisms are employed to characterize these messages, namely a standard word-based approach and a character n-gram document representation. Moreover, two distinct data sets, both with and without e-mail header information, are compared. Based on these document representations, the categorization performance of the various machine learning approaches is assessed and a comparison is given. The results indicate a substantial increase in classification accuracy when header information is considered in the document representation. To a much lesser degree, word-based document representations are advantageous over n-gram representations.