网页自动分类中特征权重计算方法的改进  

An Improved Approach to Term Weighting in Automatic Web Page Classification

作　　者：谭金波[1] 杨晓江[2] 李艺[2] 

机构地区：[1]山东师范大学教育技术系,济南250014 [2]南京师范大学教育技术系,南京210097

出　　处：《情报学报》2008年第1期56-61,共6页Journal of the China Society for Scientific and Technical Information

摘　　要：目前，在网页分类中，对HTML主要结构特征进行加权的常用方法是绝对数值加权方法。这种方法的缺点是加权系数为定值，其对长文本和短文本所起的作用不同，使得结构特征对正文的影响随着正文长度的增加而削弱。针对该缺点，本文提出了一种改进型加权方法，即相对数值加权方法。通过网页层次分类的实验，比较了这两种方法对单个标签域以及多个标签域结合的分类性能。实验结果表明，相对数值加权方法能有效提高分类的精确度，并且效果优于绝对数值加权方法。Currently, in Web page classification, absolute weighting method is the common method to weight HTML main structure features. The disadvantage of the method is that weighting coefficient is fixed value, which has different effect to long text and short text. So the influence Of structure features to body text will weaken with the length of body text increase. To solve the problem, we propose an improved weighting method, namely relative weighting method. With the experiment of Web page hierarchical classification, we compare the two methods＇ classification performance on a single label and several labels combination. The results show that relative weighting method can effectively improve the classification accuracy, and the effect is better than absolute weighting method.

关 键 词：网页分类 层次分类 特征项 权重算法 相对数值加权 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] F830.59[自动化与计算机技术—计算机科学与技术]