Abstract: The target of cluster analysis of Web log is to obtain visiting pattern of user’s interests, and provide customized personalization service for various user groups. Since the raw CURE with the weaknesses of randomness in selecting representative point and insufficiency in reflecting user interest and preference, this paper presents an improved use clustering algorithm in order to clustering user by retrieving main attributes of element with user outstanding features, and to provide initial class sets for merging small classes. The experimental results prove that it can obtain a better cluster result with this method.

Key words: Clustering Using Representative（CURE） algorithm, cluster analysis, users interests, personalization

摘要： 通过对Web网站的日志进行聚类分析，目的是获取用户兴趣访问模式，进而为不同用户群体提供定制的个性化服务。针对原始CURE算法在代表点选择的随机性、不能充分体现用户兴趣偏好方面存在的问题，提出了改进的用户聚类算法，根据用户兴趣的显著特征提取元素的主要属性进行预聚类，为小类合并提供合理的初始类集，实验结果证明了该方法有较好的聚类结果。

关键词: 利用代表点聚类（CURE）算法, 聚类分析, 用户兴趣, 个性化