分类属性数据量子聚类算法的改进  

IMPROVING QUANTUM CLUSTERING ALGORITHM OF CATEGORICAL DATA

作　　者：赵正天[1] 赵小强[1] 李炜[1] 段晓燕[2] 

机构地区：[1]兰州理工大学电气工程与信息工程学院,甘肃兰州730050 [2]兰州石化职业技术学院电子电气工程系,甘肃兰州730060

出　　处：《计算机应用与软件》2010年第12期101-104,共4页Computer Applications and Software

基　　金：甘肃省自然科学基金(3ZS051-A25-032);甘肃省教育厅高等学校研究生导师科研项目(050301)

摘　　要：分析量子势能、量子力学中粒子分布机制和分类属性数据的量子聚类CQC(Categorical Quantum Clustering)算法。针对CQC算法存在的聚类效果对聚类度量尺度β较敏感,而β的选取往往凭经验确定没有通用原则,以及对线性可分数据聚类效果显著,但对线性不可分数据不能奏效等问题,通过引入新的相异性度量测度及聚类度量尺度步长βstep,重新定义紧致性指标ICD,提出一种改进的ICQC算法。该算法首先在不同粒度水平上划分数据样本产生初始类(簇),之后采用聚类中心间相异性测度最近邻方法合并初始类(簇)完成聚类。通过与CQC算法的实验比较,证明该算法具有更高的聚类效能,在CQC算法失效的情况下,也能获得良好的聚类效果。In this paper we analysed quantum potential,distribution mechanism of particles in quantum mechanics and CQC algorithm of quantum clustering of categorical data. The CQC algorithm has problems,including the clustering effect is sensitive to cluster metrics scale β which often depends on the experience to be determined when selecting but has no general principle,and being remarkable in effect to linear separable data but ineffective to linear inseparable data,etc. To resolve these,an improved ICQC （ Iterative categorical quantum clustering） algorithm is proposed by introducing new dissimilarity metrics measure and clustering metrics scale step βstep and redefining compactness index ICD. The improved ICQC partitions firstly the data sample in different granularities levels to arise initial category （ cluster） ,and then mergers the initial category （ cluster） to complete clustering by using the nearest neighbour method of dissimilarity metrics among clusters’centres. Through the experimental comparison with CQC algorithm,the higher clustering performance of the proposed algorithm is attested. Even when the CQC algorithm fails,the proposed algorithm can still acquire favourable effect of clustering.

关 键 词：分类属性 量子聚类 相异性度量测度 聚类度量尺度步长 紧致性指标 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]