摘要
针对传统K-means算法因初始聚类中心的随机性而导致聚类结果产生很大的波动性问题,提出一种基于最小距离乘积聚类算法CAMDP(Clustering Algorithm based on Min-Distance Product),利用数次抽样技术,在得到的聚类中心集合上继续使用最小乘积法寻找最佳的初始聚类中心,较大程度减少了K-means聚类算法对初值选取的随机性。实验结果表明:改进后的K-means算法既考虑了网络结构的拓扑信息,又考虑了节点的属性特征,为社区划分提供了有力的决策支持。
Traditional K-means algorithm of the initial clustering center is randomly generated,which can lead to produce very big volatility clustering results. In order to solve this problem,We propose a algorithm named clustering algorithm based on min-distance Product. With the method of sampling,CAMDP（ Clustering Algorithm based on Min-Distance Product） produces selected point which has minimum product of distances between itself and all other initialized clustering centers,which improves the selecting of the initial value of the K-means algorithm,avoiding the random selected clustering centers. The results show that the topological feature is considered and the attributes of vertex are taken into account,which let the improved K-means provide the strong support to the division of community.
出处
《吉林大学学报（信息科学版）》
CAS
2015年第5期564-569,共6页
Journal of Jilin University（Information Science Edition）
基金
国家青年自然科学基金资助项目(61300145)
关键词
社区结构
聚类
社会关系
聚类中心
community structure
clustering
social relations
clustering centers
