混合约束的软限制近邻传播半监督聚类算法

Hybrid Constrained Semi-Supervised Clustering Algorithm Based on Soft-Constraint Affinity Propagation

提出了一种混合约束的半监督聚类算法HCSCAP,综合考虑了已标号点和成对点约束信息,使2类先验信息在聚类的过程中能发挥各自的作用.通过调整相似性矩阵添加成对点约束,已标号点以宏结点的方式添加到相似性矩阵.给出了具体的算法步骤并进行了测试,实验表明:HCSCAP比只利用成对点约束信息的SAP算法和只利用标号点的SS-CAP算法的CRI指标要好,聚类簇数也更接近实际给定的类数.

A hybrid constrained semi-supervised clustering algorithm（HCSCAP） is proposed on the base of soft-constraint affinity propagation algorithm.In order to get a better clustering result,both labeled data and pair-wise constraints are considered in clustering to make use of two types of prior knowledge supplementary to each other.We exploit pair-wise constraints by adjusting the similarity matrix,and append labeled data as macro-nodes to the similarity matrix.The experiments show that the performance of HCSCAP is better than that of SAP which makes use of pair-wise constraints only and that of SSCAP which makes use of labeled data only.The number of clusters given by HCSCAP is more close to the actual class number.