一种基于语料特性的聚类算法  

Clustering Algorithm Based on the Distributions of Intrinsic Clusters

作　　者：曾依灵[1,2] 许洪波[1] 吴高巍[1] 白硕[1] 

机构地区：[1]中国科学院计算技术研究所网络重点实验室,北京100190 [2]中国科学院研究生院,北京100049

出　　处：《软件学报》2010年第11期2802-2813,共12页Journal of Software

基　　金：国家自然科学基金No.60933005;国家重点基础研究发展计划(973)Nos.2007CB311100;2004CB318109;国家高技术研究发展计划(863)No.2007AA01Z441~~

摘　　要：为寻求模型不匹配问题的一种恰当的解决途径,提出了基于语料分布特性的CADIC(clustering algorithm based on the distributions of intrinsic clusters)聚类算法.CADIC以重标度的形式隐式地将语料特性融入算法框架,从而使算法模型具备更灵活的适应能力.在聚类过程中,CADIC选择一组具有良好区分度的方向构建CADIC坐标系,在该坐标系下统计固有簇的分布特性,以构造各个坐标轴的重标度函数,并以重标度的形式对语料分布进行隐式的归一化,从而提高聚类决策的有效性.CADIC以迭代的方式收敛到最终解,其时间复杂度与K-means保持在同一量级.在国际知名评测语料上的实验结果表明,CADIC算法的基本框架是合理的,其聚类性能与当前领先水平的聚类算法相当.In finding a flexible approach to solve the model misfit problem,a clustering algorithm based on the distributions of intrinsic clusters（CADIC） is proposed,which implicitly integrates distribution characteristics into the clustering framework by applying rescaling operations.In the clustering process,a set of discriminative directions are chosen to construct the CADIC coordinate,under which the distribution characteristics are analyzed in order to design rescaling functions.Along every axis,rescaling functions are applied to implicitly normalize the data distribution such that more reasonable clustering decisions can be made.As a result,the reliability of clustering decisions is improved.The time complexity of CADIC remains the same as K-means by using a K-means-like iteration strategy.Experiments on well-known benchmark evaluation datasets show that the framework of CADIC is reasonable,and its performance in text clustering is comparable to that of state-of-the-art algorithms.

关 键 词：CADIC(clustering algorithm based on the DISTRIBUTIONS of INTRINSIC clusters) 文本聚类 模型不匹配 重标度 信息检索 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]