A New Approach to the Multiaspect Text Categorization by Using the Support Vector Machines

Abstract

In our earlier work we introduced the concept of the multiaspect text categorization (MTC) task which has its roots in relevant practical problems of managing collections of documents at many, if not all, commercial companies and, above all, public institutions. Specifically, it is a well defined general problem which boils down to the classification of textual documents at two levels: first, to a general category, and—second—to a specific sequence of documents within such a category. While the former task may be dealt with the use of some standard text categorization techniques, the latter one is more challenging due to, first of all, a limited number of training documents. On the other hand, it is assumed that there is some natural logic, for instance, resulting from rules and regulations, behind the succession of documents within the sequences which can be exploited to make a decision as to the assignment of a new document to a proper sequence. We have studied the MCT problem in a number of papers and proposed some solutions to it. Here we propose a new solution which is based on the use of the support vector machines (SVMs) which are known as a very effective technique to solve various classification tasks. We consider the application of SVMs in a specific context, determined by the characteristics of the MTC problem, and by a specific data set used for the experimentation. The use of the SVMs has implied a new, more sophisticated representation of the documents and their sequences which has made it possible to obtain promising results in computational experiments. Moreover, the proposed approach is flexible and may be considerably modified and extended to cover many possible problem versions.