基于粗糙集和特征位置重要度的特征加权方法  

Feature Weighting Method Based on Rough Sets and Feature Positions Important Degrees

作　　者：王勋[1] 裴志利[2] 王庆虎[2] 

机构地区：[1]内蒙古民族大学数学学院,内蒙古通辽028043 [2]内蒙古民族大学计算机科学与技术学院,内蒙古通辽028043

出　　处：《内蒙古民族大学学报（自然科学版）》2013年第2期150-153,共4页Journal of Inner Mongolia Minzu University：Natural Sciences

基　　金：国家自然科学基金资助项目(61163034);内蒙古自治区人才基金资助项目(2011);吉林大学符号计算与知识工程教育部重点实验室开放课题基金资助项目(93K172011k07);科技部"十一五"国家科技支撑计划项目(2009BH41B03)

摘　　要：文本预处理是文本分类的首要任务,特征加权直接影响文本预处理的质量,本文提出了基于粗糙集和特征位置重要度加权的方法,该方法的主要思想是:在计算传统TF-IDF特征权重的过程中将已有的决策信息和特征的位置信息引入到权重计算中,考虑可变精度粗糙集模型中的近似分类质量和近似分类精度以及特征位置信息对全局分类的作用.经过多重文本分类实验证明,本文提出的特征加权方法能够明显的提高文本分类的效率.Text pretreatment is the primary task of text categorization,the result of text pretreat-ment was directly affected by feature weighting,we presented feature weighting method based on rough sets and feature positions important degrees in this paper,the main idea of this method is：The existing decision information and the location information of feature is introduced into the calculation of the weight in the calculation process of traditional TF-IDF feature weight,Considering the role of approximation categorization quality and approximation categorization accuracy in variable accuracy rough set model as well as the feature position information for the global categorization.After multiple text categorization experiments show,the feature weighting method in this paper,can significantly improve the performance of text categorization.

关 键 词：近似分类质量 近似分类精度 位置信息 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]