Constructing feature set by using temporal clustering of term usages in document categorization

Abstract

For discovering some chances in documents with temporal context, it is important to handle their contents represented as words and phrases, called keywords. However, in conventional methods, keywords are selected based on their frequency and/or a particular importance index such as tf-idf throughout their observed period. In this chapter, we describe a method for characterizing large number of documents, considering the temporal features of appeared terms, by obtaining document clusters based on the similarities between the document that are characterized by the temporal patterns of an importance index for considering temporal differences in term usages. As an experiment, we performed document clustering for four sets of bibliographical documents using two feature sets: popular feature terms appearances and the appearances of temporal patterns for each document. Then, we compared the time dependencies of the two document clustering results. Our feature construction method succeeded in representing the time differences in the documents using features based on temporal patterns.