Integrating Linguistic Resources in TC through WSD

Abstract

Information access methods must be improved to overcome theinformation overload that most professionals face nowadays. Textclassification tasks, like Text Categorization, help the usersto access to the great amount of text they find in the Internetand their organizations.TC is the classification of documents into a predefined set ofcategories. Most approaches to automatic TC are based on theutilization of a training collection, which is a set of manuallyclassified documents. Other linguistic resources that areemerging, like lexical databases, can also be used forclassification tasks. This article describes an approach to TCbased on the integration of a training collection (Reuters-21578)and a lexical database (WordNet 1.6) as knowledge sources.Lexical databases accumulate information on the lexical items ofone or several languages. This information must be filtered inorder to make an effective use of it in our model of TC. Thisfiltering process is a Word Sense Disambiguation task. WSDis the identification of the sense of words in context. This taskis an intermediate process in many natural language processingtasks like machine translation or multilingual informationretrieval. We present the utilization of WSD as an aid for TC. Ourapproach to WSD is also based on the integration of two linguisticresources: a training collection (SemCor and Reuters-21578) and alexical database (WordNet 1.6).We have developed a series of experiments that show that: TC andWSD based on the integration of linguistic resources are veryeffective; and, WSD is necessary to effectively integratelinguistic resources in TC.