一种应用复杂网络特征的K-means初始化方法

K-means initialization method using properties of complex network

K-means算法是一种基于划分的聚类算法,具有算法简单且收敛速度快的特点。但该算法的性能依赖于聚类中心的初始位置的选择。拓展了复杂网络的重要特征,针对带有属性的数据对象所构成的数据集,定义了多维属性对象的度、聚集度和聚集系数,选取度和聚集系数高的K个点作为K-means聚类的初始中心点。实验数据表明,改进后的K-means算法较传统的算法具有更高的效率和准确度。

K-means algorithm is a partition-based clustering algorithm.It is simple and fast to converge,the performance of K- means algorithm depends on that how to choose K samples as the initial cluster centers.This paper develops the properties of complex network,and defines degree,congregated degree and congregated coefficient of objects with feature,and chooses the K nodes whose the degree and congregated coefficient are larger than the others as the initial cluster centers.The experiment shows that the improved K-means clustering algorithm is more efficient than the original K-means clustering algorithm.