摘要： 近邻传播算法在非凸形、密度不均匀的数据集上很难得到理想的聚类结果。为此，基于核聚类的思想，将数据集非线性地映射到高维空间，使数据集更加分离。利用共享最近邻的相似度度量方法，提出一种密度不敏感的近邻传播算法DIS-AP，以弥补原算法易受特征集维数和密度影响的缺点，从而有效解决数据集非凸和密度不均匀问题，拓宽算法的应用范围。仿真实验结果证明，DIS-AP算法具有更好的聚类性能。

关键词: 近邻传播, 相似度度量, 核聚类, 共享最近邻, 聚类分析, 密度不敏感

Abstract: To solve the problem that Affinity Propagation(AP) algorithm has poor performance on non-convex and asymmetrical density dataset, kernel clustering is introduced into algorithm. The dataset in kernel space are farther separable through non-linear mapping. Then a similarity measure with shared nearest neighbor is imported, and a density insensitive-affinity propagation algorithm named Density-insensitive Affinity Propagation(DIS-AP) is proposed. DIS-AP overcomes the shortcoming of original AP based on Euclidean distance that is easily influenced by the dimension and density of dataset. It can effectively solve the problem of clustering non-convex and asymmetrical density dataset, and developed its applied range. Experimental results show that this algorithm has better clustering effect.

Key words: Affinity Propagation(AP), similarity measurement, kernel clustering, shared nearest neighbor, clustering analysis, density insensitive

中图分类号: 

• TN915