Rearranging classified items in hierarchies using categorization uncertainty

Abstract

The classification into hierarchical structures is a problem of increasing importance, e.g. considering the growing use of ontologies or keyword hierarchies used in many web-based information systems. Therefore, it is not surprising that it is a field of ongoing research. Here, we propose an approach that utilizes hierarchy information in the classification process. In contrast to other methods, the hierarchy information is used independently of the classifier rather than integrating it directly. This enables the use of arbitrary standard classification methods. Furthermore, we discuss how hierarchical classification in general and our setting in specific can be evaluated appropriately. We present our algorithm and evaluate it on two datasets of web pages using Naïve Bayes and SVM as baseline classifiers.