权值矩阵聚类算法

A Weight Matrix Clustering Algorithm

由于以往的算法不能对用户感兴趣的页面进行很好的聚类,所以将网站访问频度作为参数考虑进来,提出一个新的概念——权值关联矩阵,以Web服务器URL为行、以UserID为列建立URL-UserID关联矩阵,与普通的矩阵聚类算法相比,根据用户对某页面的兴趣度,再生成权值关联矩阵。从而发现相似的用户群体和相似的W eb页面。该算法通过上机实践,与传统的矩阵聚类算法相比具有识别准确率高,用户向量特征描述更准确,且能够更准确的反映网站的访问情况等优点。同时为用户提供个性化推荐服务铺平了道路。

Because the algorithms can not cluster the interested pages well, the visit frequency of the site is taken as a parameter into account to introduce a new concept - the right of correlation matrix, namely a URL - UserID correlation matrix. Compared to the matrix clustering algorithm, the right correlation matrix is generated according to the pape on users interest. Thus a similar user groups and similar Web pages are found. The algorithm matrix has higher identification accuracy, more accurate description of vector features, and can more accurately reflect the site visits compared with traditional clustering algorithms.