基于改进近邻传播算法的Web用户聚类  

Clustering of Web Users Based on an Improved Affinity Propagation Algorithm

作　　者：冷亚军[1,2] 梁昌勇[1,2] 陆文星[1] 

机构地区：[1]合肥工业大学管理学院,合肥230009 [2]过程优化与智能决策教育部重点实验室,合肥230009

出　　处：《情报学报》2012年第9期993-997,共5页Journal of the China Society for Scientific and Technical Information

基　　金：基金项目：国家自然科学基金重点项目（70631003）.

摘　　要：随着Intemet和电子商务的迅猛发展，聚类技术在Web用户划分方面的作用越来越明显。Web用户聚类的难度在于有成千上万的用户需要聚类，而且每个用户的偏好向量是高维稀疏的。对于处理大规模的数据集。近邻传播算法是一种快速、有效的聚类方法。但面对高维稀疏的数据，近邻传播算法往往不能得到很好的聚类结果，而且该方法不能产生指定类数的聚类。本文提出一种改进的近邻传播算法，使用该方法对Web用户进行聚类。根据灰关系等级和Jaccard系数定义用户相似度矩阵，对算法产生的初始聚类进行重新分配，获得指定类数的聚类。实验结果表明新算法是有效的，与原始近邻传播算法相比，新算法在个性化推荐的应用中具有更好的性能。With the rapid development of Internet and e-commerce, clustering becomes more and more important for grouping Web users. Web user clustering is difficult because thousands of users may have to be clustered, and also because each user＇s preference vector is high-dimensional and sparse. Affinity propagation algorithm is an efficient and fast clustering method for large datasets compared with the existing clustering algorithms. But for the high-dimensional and sparse datasets, affinity propagation algorithm can not produce good clustering results. Another limitation of affinity propagation algorithm is that it can not produce clustering results with given number of clusters. In this paper, an improved affinity propagation algorithm is proposed for Web user clustering. Grey relational analysis and Jaccard coefficient are applied to compute the user similarity matrix. Initial clusters produced by the proposed algorithm are redistributed to generate the clustering results with given number of clusters. Experimental results show that the proposed algorithm is efficient, it performs better than the original affinity propagation algorithm in the application of personalized recommendation.

关 键 词：WEB用户聚类 稀疏性 近邻传播算法 相似度矩阵 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]