二维空间中硬聚类算法影响力因子的作用研究  

Study on Influence of Effectiveness Factor in HCM Algorithm in 2-D Space

作　　者：金健[1] 黄国兴[1] 梁道雷[1] 

机构地区：[1]华东师范大学信息科学技术学院,上海200062

出　　处：《计算机科学》2006年第10期182-185,共4页Computer Science

基　　金：华东师范大学211重点项目资助。

摘　　要：经典硬聚类算法HCM(hardc-means)完全基于欧氏距离,针对其无法较好应对各簇规模差异较大的情况,提出在每个欧氏距离项上加入一个影响力因子,使基于距离的标准转变为更通用的基于角度的标准的方法(HCMef算法)。用该算法对二维空间中两类分布密度基本一致,样本数对比分别为1000∶1000、1000∶5000和1000∶10000,正态分布且类边界从较模糊到较清晰的不同数据进行试验。结果显示,HCMef方法可以很好地找到聚类中心的标准设定值,在各种情况下都有很明显优势,表现出很强的稳定性。表明该方法在二维两类情况下的可行性,并值得做进一步推广研究。The classic hard c-means （HCM） is totally based on Euclid Distance, and it cannot cope with situations of different cluster sizes. Method of attaching an effectiveness factor to each distance item （HCMef） is proposed, transforming the criterion based on distance into the more general criterion based on angle. Two-cluster data sets in 2-D space, normally distributed, with the similar density, the cluster boundaries from vague through clear, and the contrast of cluster population 1000 ： 1000, 1000 ： 5000 and 1000 ： 10000 respectively, are experimented. The results show that HCMef can find the pre-established cluster centers faster and more precisely. The advantages of HCMef are obvious under various situations. The feasibility of HCMef in 2-D space with 2 clusters are verified and the further study is worth performing.

关 键 词：HCM 聚类 影响力因子 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]