基于有向带权图的页面聚类算法研究  

Study on Page Clustering Algorithms Based on Weighted Directed Graph

作　　者：方杰[1] 张结魁[1] 周军[1] 

机构地区：[1]合肥工业大学,安徽合肥230009

出　　处：《计算机技术与发展》2009年第9期49-53,共5页Computer Technology and Development

基　　金：国家自然科学基金项目(70672097);国家自然科学基金重点项目(70631003)

摘　　要：聚类算法是数据挖掘中的一个重要的分析工具。Web使用挖掘中的聚类分析一般分为用户聚类和页面聚类。其中页面聚类是指导网站结构离线优化的重要方法。利用有向带权图表示用户的访问会话记录,对建立的有向带权图模型运用聚类算法实现页面聚类。选取真实数据对典型的聚类算法K-means算法、DBSCAN算法和COBWEB算法进行实验。实验结果表明,在选取的数据集范围内,COBWEB算法准确率要高于K-means算法和DBSCAN算法,时间性能与用户访问频率矩阵大小有密切关系。Clustering algorithm is an important analytical tool in data mining. Clustering analysis is generally fallen into user clustering and page clustering in Web usage mining. Page clustering is an important methods for guiding for the structure of the site off- line optimization. This paper use weighted directed graph to describe user visit and conversation records, and use clustering algorithms to realize the page clustering by the weighted directed graph mode established. Select the real data carries on the experiment to the typical clustering algorithms K-means algorithm,DBSCAN algorithm and COBWEB algorithm. The experiments results indicate that in the selected data sets, the accuracy rate of COBWEB algorithm is higher than that of K - means algorithm and DBSCAN algorithm, and the time capability is closely related to the size of user visit frequency matrix.

关 键 词：有向带权图 聚类算法 页面聚类 K-MEANS算法 DBSCAN算法 COBWEB算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]