近邻传播半监督聚类算法的分析与改进

Analysis and improvement of semi-supervised clustering algorithm based on affinity propagation

近邻传播半监督聚类算法SAP在小数据集上运行时可能会出现并列类代表点的现象,当出现并列类代表点时,依据决策矩阵E对角线上数值大于0确定的类代表点并不是全部的类代表点。分析了近邻传播算法的性质,找出了并列类代表点的出现原因,并针对此现象给出了改进算法。

Parity exemplars often appear when Semi-supervised clustering algorithm based on Affinity Propagation（SAP） is applied on small dataset,and then the exemplar judgment criterioni,.e.an exemplar xk must satisfy E（k,k）0 in the decision matrix Ei,s not complete.In this paper,properties of affinity propagation algorithm are analyzed and the occurrence reason of parity exemplars is found.Finallya,n improved algorithm is proposed to solute this problem.