Using unlabeled data to improve classification in the naive bayes approach: Application to web searches

Abstract

This paper introduces a method to build a classifier based on labeled and unlabeled data. We set up the EM algorithm steps for the particular case of the naive Bayes approach and show empirical work for the restricted web page database. Original contributions includes the application of the EM algorithm to simulated data in order to see the behavior of the algorithm for different numbers of labeled and unlabeled data, and to study the effect of the sampling mechanism for the unlabeled data on the results.