基于优选链接的中文网页分类方法研究  

Research on Chinese Web Pages Classification Based on Preferential Links

作　　者：陈胜荣[1] 董守斌[1] 

机构地区：[1]华南理工大学计算机科学与工程学院,广州510640

出　　处：《郑州大学学报（理学版）》2007年第2期78-82,共5页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学资金资助项目;编号90412015;60603022

摘　　要：网页之间的链接为Web数据挖掘提供了丰富信息,通过链接关系来加强中文网页分类的效果.由于网页的编写存在随意性和不规范性,并非所有的链接页面与源页面都有主题相关性.通过比较页面与其链接页面的相似性,优选出相似度较高的链接子集,利用得分向量对基于内容的分类结果进行修正.基于北大天网数据集的实验结果表明,优选链接页面的分类方法对于训练样本较少的类别在分类精度上有一定的提高.Links among pages provide rich information for Web mining. Links relationship improve the Chinese Web pages classification. As Web pages are created casually, not all linked pages share the same topic as the source page. So, by comparising the similarity between page and its linked pages, a subset of the linked pages that has the highest similar score is chosen and the original classification results are modified by score vector. The experiment based on the Chinese dataset provided by Tian Wang of Peking University shows that this method improves classification in recall, precision as well as F1 value.

关 键 词：分类 链接 相似度 得分向量 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]