A Hybrid User Model in Text Categorization

A user model that specifies user preferences on message handling is an essential component of an e-mail message categorizer. We present an approach that combines two learning algorithms, i.e. the Naïve Bayesian Classifier (NBC) and Progol, to model implicitly and explicitly reflected user preferences that may not be modeled by using either the algorithms alone. An experiment demonstrates the improvement of categorization performance compared to that of using the two algorithms independently.