一种基于混合数据相似性度量的谱聚类算法  

Spectral Clustering Algorithm Based on Mixed Data Similarity Measure

作　　者：马恒[1] 丁世飞[1,2] 

机构地区：[1]中国矿业大学计算机科学与技术学院,徐州221116 [2]中国科学院计算技术研究所智能信息处理重点实验室,北京100190

出　　处：《小型微型计算机系统》2016年第8期1746-1750,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61379101)资助;国家"九七三"重点基础研究发展计划项目(2013CB329502)资助

摘　　要：随着科技的发展,人们在生活中产生了大量的数据,其中部分数据具有数值型和分类型两种属性类型.现有的大多数聚类算法只能处理单一属性类型的数据,对这种混合属性的数据往往难以处理.针对这个问题提出一种基于混合数据相似度测量的谱聚类算法,首先对两种属性数据分别进行相异度度量,然后用一种相似性度量表示出混合数据之间的相似性关系,把相似性关系映射成无向图两顶点之间边的权值,最后通过谱聚类算法实现聚类划分.从UCI标准数据集选取几个混合数据集进行实验,并与其他算法进行了比较,验证了本算法对混合数据聚类的有效性.With the development of the technology, people have produced a large amount of data in the life, some of these data with numerical and categorical two types of attributes. Most existing clustering algorithms can only deal with the data has one type attribute, they are often difficult to deal with mixed attribute data. To solve this problem,this paper proposes a spectral clustering algorithm based on mixed data similarity measure, First build the dissimilarity measure for two kind of attribute data respectively, Then base on the dissimilarity measure, establish the similarity relationship between the mixed data with a similarity measure, let one mixture data as an undirected graph vertex,the similarity relations between the mixed data are mapped into the undirected edge weight between two vertices, Finally through Spectral clustering to achieve clustering of mixed data. Selected several mixed data set for experiment from UCI standard data sets, and compared with other clustering algorithm for mixed data, Verified this algorithm is effective for mixed data clustering.

关 键 词：混合数据 谱聚类 聚类 相似度量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]