在汲取了传统划分、层次聚类方法优点的基础上,结合图搜索技术,提出了一种新的快速混合聚类算法.该算法主要分为三个步骤:首先将整个数据集“压碎”,生成固定数量的原子级聚类;然后处理孤立点;最后采用图搜索技术生成聚类.该算法只需一个参数,能识别任意形状、大小的聚类,时间复杂度在最坏情况下为n√n.实验表明该算法是有效的.
Combing with graph-based searching technique and the advantadges of the traditional partition and hierarchical clustering methods, SHILL (a new speedy hy'brid clustering algorithm) is proposed. SHILL contains three steps: first, crush the whole data set into a number of atom-class clusters; second, deal with isolated point; then, adopt graph-based searching technique to create clusterings. SHILL only requires one parameter, and can discover arbitrary shapes and sizes of cluster, and the time complexity of SHILL is n√n under the worst condition. Experimental results show that the algorithm is effective.