Two-Stage Text Classification Using Bayesian Networks

The“curse of dimensionality”provides a powerful impetus to explore alternative data structures and representations for text processing. This paper presents a method for preparing a dataset for classification by determining the utility of a very small number of related dimensions via a Discriminative Multinomial Naive Bayes process, then using these utility measurements to weight these dimensions for use in a Bayesian network classifier. We show that the use of this two-stage methodology provides significant improvements over both Discriminative Multinomial Naive Bayes and Bayesian network classifiers alone.