基于特征权重优化的改进KNN Web文本分类算法  

An Improved KNN Algorithm Applied to Web Text Categorization Based on Weight Optimization

作　　者：王煜[1,2] 白石 王正欧[2] 

机构地区：[1]河北大学数学与计算机学院,保定071002 [2]天津大学系统工程研究所,天津300072 [3]沧州市城建档案馆,沧州061000

出　　处：《情报学报》2007年第5期643-647,共5页Journal of the China Society for Scientific and Technical Information

基　　金：国家自然科学基金资助项目（60275020）.

摘　　要：本文提出了一种基于权重优化的样本相似度测量的距离公式,改进了KNN文本分类算法.KNN算法通常采用传统的VSM模型,各个特征具有相同的权重,使其不适应于文本处理的环境.本文首先根据神经网络理论,采用灵敏度方法对文本特征向量的每个特征的权重进行修正,并且采用降低运算量的神经网络特征选择方法进行第二次降维处理.然后根据同一特征对不同类别的文本类的分类作用不同,对距离公式中的特征权重进行进一步改进,从而进一步提高了KNN文本分类算法的精度.In this paper, an improved KNN method applied to text categorization is proposed, which is based on weight optimization of text features. When the KNN method is applied to text categorization, the traditional VSM model is not suitable for the situation when all features of the text vector have the same weight. In this paper, based on the neural network theory, weights of features are adjusted firstly by using sensitivity method. And the second features selection is done again by using the neural network, which has decreased the work of computing. Then, based on different function of the same feature effect on different class, the distance metric is improved further. Subsequently, the text categorization accuracy of KNN is improved.

关 键 词：文本分类 神经网络 KNN算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]