方向相似性聚类方法DSCM

The Directional Similarity-Based Clustering Method DSCM

针对方向性数据提出了一种鲁棒的基于方向相似性度量的聚类方法DSCM·DSCM首先基于方向性度量构造目标函数,然后通过不动点迭代法对目标函数优化,获得各个样本的最终稳定状态,最后基于样本的最终状态集利用层次聚类技术实现聚类·DSCM的优势在于对方向性数据聚类时不依赖于具体的初始化参数,且能自组织地求解最优聚类划分因而有很好的鲁棒性·通过实验证实了DSCM的有效性以及对已有的两个传统方向性聚类算法的优越性·

The directional similarity-based robust clustering approach DSCM for directional data is presented in this paper. The DSCM utilizes the objective function based on the directional similarity measure, and then optimizes it using the fixed point iteration method such that all the stable states of the data samples are derived. By presenting all these stable states to the hierarchical clustering algorithm AHC, the final clustering results are obtained. This new approach exhibits its robustness to initialization and capability to reasonably detect the number of clusters for clustering directional data. The experimental results demonstrate its validity and distinctive superiority over the two conventional directional clustering algorithms.