维度概率摘要模型及其层次聚类算法

Hierarchical clustering algorithm with dimensions probability summary model

提出一种维度概率摘要模型,将聚类产生的簇摘要信息采用各维度的概率分布来表示;定义点簇相似度、簇簇相似度等相似性度量方法;提出一种基于维度概率摘要模型的凝聚层次聚类算法.实验分析发现,所提模型和算法能够产生高质量的聚类,能够避免噪声点的影响并发现离群点,能够自动发现聚类,算法稳定可靠且对高维数据集聚类效果很好.

A dimensions probability summary model is proposed, which uses dimensions probability distributions to represent the cluster summary. Similarities from point to cluster and from cluster to cluster are defined as the similarity measurement. A hierarchical clustering algorithm based on the dimensions probability summary model is proposed. Experimental analysis shows that, the model and algorithm can produce high quality of the clustering, avoid the noise and find the outliers, and automatically determine the cluster number. The proposed algorithm is stable and reliable, and is suitable for high dimensional data clustering.