一种基于特征重要度的文本分类特征加权方法.pdf

ISSN 1000- 1239PCN 11- 1777PT P Journal of Comput er Research and De elopment 46( 10) : 1693- 1703, 2009 1, 2 1, 2 3 1, 2 刘 赫 刘大有 裴志利 高 滢 1 ( 1300 12) 2 ( 1300 12) 3 ( 028043) ( liuhe1980@ 163. com) A Feature Weighting Scheme for Text Categorization Based on Feature Importance 1, 2 1, 2 3 1, 2 Liu He , Liu Dayou , Pei Zhili , and Gao Ying 1 (College of Comp uter S cience and Technology , J ilin Univers ity, Chang chun 1300 12) 2 ( M inistry of Education K ey L a oratory of Sy m olic Comp utation and K now led ge Engineering , J il in University , Changchun 130012) 3 (College of Comp uter Science and T echnology , I nner M ongolia Univ ers ity f or N ationalities, T ongl iao, I nner M ongolia 028043) Abstract T ext categorization is one of the key research fields in text mining. Feature w eighting is an important problem in text categorization. For computing feature w eights, a feature w eighting scheme for text categorization is proposed. In this scheme, the feature importance is defined based on the real rough set theory. By this concept, decision-making information of a featur e for categor ization is introduced into the w eight of this feature. T hen, the experiments are perfor med on tw o international and standard text datasets, namely, Reuters-21578 Top 10 and WebKB. Through the computation of the total w ithin-class scatter and betw een-class scatter in Fisher linear discriminant, it is erified that the proposed scheme can decrease the total w ithin-class scatter and increase the betw een-class scatter; that is to say , the scheme can make samples in the same class mor e compact and th