Feature Selection and Integration in Automatic Classification of Japanese Texts

概要

論文の詳細を見る
We explore the problem of automatic text clas-sification using Japanese documents. Unlike other languagesthat use roman letters, Japanese language poses a problem ofnot availing word boundary information. As such bag-of-wordsapproach in constructing features may not be sufficient to en-hance machine learning techniques. We propose a method forfeature selection and construction to improve automatic clas-sification performance of Japanese texts. Our approach in-volve extracting syntactic word categories and Chinese charac-ters (Kanji) separately. Then we combine the extracted infor-mation to build an informative feature set. We carried out var-ious experiments using four learning algorithms to evaluate itseffectiveness. The proposed method generally outperformed itscounterparts method for Japanese document representation.