关键词: 最大熵模型, 邮件分类, 特征, 层次分类

Abstract: Email categorization assigns new emails to pre-defined categories based on their contents and properties.In this paper，the maximum entropy model is applied to email categorization.The pre-process of email is discussed firstly，and features extracted from email header are presented.No only the effects of categorization performance coursed by Email feature fields，the numbers of features and iteration are presented and discussed，but also the performance of hierarchy categorization and direct categorization is compared.The results of experiments show that the appropriate numbers of features and iteration are 2 000，250 respectively and utilizing all email fields gets the best whole performance than others，and that the performance of legitimate by using email header and subject is best though the whole performance is worst，which is also verified in the hierarchy categorization experiments.The results also illuminate that the effect of hierarchy categorization is better than that of direct categorization.The summarization and future work are presented in the end.

Key words: maximum entropy model, e-mail classification, feature, hierarchy categorization