Constructing Blog Entry Classifiers Using Blog-Level Topic Labels

Identification of a blogger's interest is usually solved as a classification problem of a sequence of his/her blog entries. In constructing a blog entry classifier, we need as training data a rather large set of blog entries that are manually labeled with a class label. In contrast, we can easily obtain a set of blog sites with class labels. In this paper, we present a method for constructing a blog entry classifier using only a set of blog sites with class labels. Our method is based on the Naive Bayes classifier coupled with the EM algorithm.