摘要
传统的聚类算法如Kmeans等,往往需要事先定义聚类数目。在实际应用中,多基于经验知识来确定类别个数,而且一般需要多次尝试,这种方法具有很大的盲目性。本文提出一种基于SOM的聚类算法,利用SOM的可视化功能和人眼在低维情况下对模式的快速识别能力来避免传统聚类算法确定聚类数目的盲目性。将提出的方法应用于某电信公司客户分群的实际问题当中,来刻画客户组的个性行为特征,以便销售人员制定针对性的营销策略,具有重要的实际意义。
The typical cluster algorithms, such as Kmeans, mostly need to decide the number of clusters before training. However,it is difficult to achieve this in fact, and it is of blindness to do this based on experience. In this paper, a clustering algorithm based on SOM is presented. SOM has powerful visualization ability, and our eyes can exploit models under low-dimension circumstances quickly. The method we present here makes use of these two advantages. Also we partition telecom customers with this method for the sake of grasping the characteristics of customer clusters. The application is meaningful in making the best telecom policy.
出处
《计算机工程与科学》
CSCD
2007年第8期133-136,共4页
Computer Engineering & Science
基金
国家自然科学基金资助项目(60421002)
关键词
数据挖掘
聚类
SOM
可视化
客户分群
data mining
cluster
SOM
visualization
customer partitioning