基于流形距离的半监督近邻传播聚类算法

摘要

通过学习数据集的低维流形结构,给出一种流形距离测度；结合成对约束信息,调整数据的相似度矩阵,将其作为近邻传播算法的输入,提出了基于流形.距离的半监督近邻传播聚类算法(SAP-MD).通过在UCI标准数据集上的仿真实验表明,SAP-MD算法相比于仅利用成对约束信息的聚类算法,在聚类性能上有很大提高.%This paper studied the manifold structure to propose a manifold distance, which used to adjust the similarity matrix combined with the pairwise constraints. It took the modified matrix as the inputs of affinity propagation algorithm. And put forward a semi-supervised affinity propagation based on manifold distance algorithm (SAP-MD). The promising experimental results on the UCI datasets prove that the SAP-MD is better than others algorithms only incorporating pairwise constraints.