Category Levels in Hierarchical Text Categorization

Category Levels in Hierarchical Text Categorization Stephen D'Alessio Keitha Murray Robert Schiaffino Department of Computer and Information Science Iona College New Rochelle, N.Y. 10801, USA sdalessio@iona.ed u, kmurray@iona.edu, rschiaffino@iona.ed u Aaron Kershenbaum Department of Computer Science Polytechnic University Hawthorne, N.Y. 10532, USA akershen@duke.poly.edu Abstract We consider the problem of assigning level numbers (weights) to hierarchically organized categories during text categorization. These levels control the ability of the categories to attract documents during the catego- rization process. The levels are adjusted to obtain a balance between recall and precision for each category. If a category's recall exceeds its precision, the category is too strong and its level is reduced. Conversely, a cat- egory's level is increased to strengthen it if its prelision exceeds its recall. ·' The categorization algorithm used is a su~ervised learning procedure that uses a linear classifier hewed on the category levels. We are given a set of categories: organized hierarchically. \Ve are also given a training corpus of documents already placed in one or more cat- egories. From these, we extract vocabulary, words that appear with high frequency within a given category, characterizing each subject area. Each node1s vocab- ulary is filtered and its words assigned weights with respect to the specific category. Tben, test documents are scanned and categories ranked based on the pres- ence of vocabulary terms. Documents are assigned to categories based on these rankings. We demonstrate that precision and recall can be significantly improved by solving the categorization problem taking hierarchy into account. Specifically, we show that by adjusting the category levels in a principled way, that precision can be significantly improved, from 84% to 91%, on the much-studied Reuters-21578 corpus organized in a three-level hierarchy of categories. 61 1 Introduction and Background The volume of online information has drastically in- creased with the explosive use of the Internet and online databases. Text retrieval systems employed by search engines for accessing this information have difficulty keeping pace with the growth in the amount of data that needs indexing and searching. Categorization of the original text is a method of organizing and mak- ing more efficient the retrieval task by sorting informa- tion into pre-specified 11 category bins11 that can then be queried against using natural language processing sys- terns.1 The, document categorization problem is one of as- signing newly arriving documents to categories within a given hierarchy of categories. In general1 lower level categories may be part of more than one higher level category. Moreover 1 a document may belong to more than one low-level category. While the techniques de- scribed here can be applied to this more general prob- lem1 the experiments we have conducted1 to date, have been carried out on a corpus where each document is a member of a single category and the categories form a tree rather than a more general directed acyclic graph. We limited the investigation to this more specific prob- lem in order to focus the investigation on the effect of adjusting the category level …

Content truncated to 500 words