Abstract

Hierarchical classification is an effective approach to categorization of large-scale text data. We introduce a relaxed strategy into the traditional hierarchical classification method to improve the system performance. During the process of hierarchy structure construction, our method delays node judgment of the uncertain category until it can be classified clearly. This approach effectively alleviates the ‘block’ problem which transfers the classification error from the higher level to the lower level in the hierarchy structure. A new term weighting approach based on the Least Information Theory (LIT) is adopted for the hierarchy classification. It quantifies information in probability distribution changes and offers a new document representation model where the contribution of each term can be properly weighted. The experimental results show that the relaxation approach builds a more reasonable hierarchy and further improves classification performance. It also outperforms other classification methods such as SVM (Support Vector Machine) in terms of efficiency and the approach is more efficient for large-scale text classification tasks. Compared to the classic term weighting method TF*IDF, LIT-based methods achieves significant improvement on the classification performance.

Introduction

The task of text classification is to assign a predefined category to a free text document. With more and more textual information available online, hierarchical organization of text documents is becoming increasingly important to manage the data. The research on automatic classification of documents to the categories in the hierarchy is needed.

Most of the classifiers make the decision in the same flat space. Classification performance degrades quickly with larger scale data sets and more categories, especially in terms of the classification time. On the other hand, a hierarchical classification method organizes all of the categories into a tree like structure and trains a classifier on each node in the hierarchy. The classification process begins from the root of the tree until it reaches the leaf node which denotes the final category for the document.