一种Web用户行为聚类算法

Web User Action Clustering Algorithm

提出了一种新的路径相似度系数计算方法 ,并使之与雅可比相似系数结合 ,用于计算用户访问行为的相似度 ,在此基础之上又提出了一种分析 Web用户行为的聚类算法 (FCC) ,通过挖掘 Web日志 ,找出具有相似行为的 Web用户 .由于 FCC聚类算法过滤了小于指定阈值的相似度系数 ,大大缩小了数据规模 ,很好地解决了其他聚类算法 (如层次聚类 )在高维空间聚类时的“维数灾难”问题 ,最后的实验结果很好 .

A clustering algorithm called FCC (Filter Coefficient Clustering) is proposed in this paper to get similitude action of Web users through mining web logs. First, we define a novel similarity coefficient called CM, and then combine it with the Jaccard Coefficient. As a result, a mix Coefficient is obtained to represent the similarity of Web user actions. The main reason why our algorithm is superior to other algorithms such as hierarchical clustering is that a threshold is engaged to filter the web log, thus the volume of data processed declined dramatically and the 'dimension disaster' problem is resolved. The experiment result is satisfactory.