一种基于随机游走的软子空间聚类集成方法  

A Method for Soft Subspace Clustering Ensemble Based on Random Walk

作　　者：李嫚 王立宏 LI Man;WANG Li-hong(School of Computer and Control Engineering,Yantai University,Yantai 264005,China)

机构地区：[1]烟台大学计算机与控制工程学院,山东烟台264005

出　　处：《烟台大学学报（自然科学与工程版）》2022年第3期343-350,370,共9页Journal of Yantai University(Natural Science and Engineering Edition)

基　　金：国家自然科学基金资助项目(62072391)。

摘　　要：聚类集成旨在将多个基聚类进行组合,以获得性能更佳、鲁棒性更强的共识结果。本文提出了一种基于随机游走的聚类集成方法CERW(Clustering Ensemble based on Random Walk)。该方法以hedonic games(享乐博弈)生成的簇为图节点,簇之间的社会福利值为边权重构造簇级相似图。为了探索相似图中节点之间的结构信息,定义一个转移概率矩阵,并在此基础上进行随机游走,通过分析不同节点的随机游走轨迹来发现博弈簇之间的潜在关系。在多个数据集上的实验结果证明了所提方法的有效性。Clustering ensemble aims at combining multiple base clusterings to obtain a better-quality and more robust consensus results. A new mothod based on random walk named CERW(Clustering Ensemble based on Random Walk) is proposed. With the clusters generated by hedonic games as nodes, and the social welfare values of clusters as edge weights, we construct a cluster-level similarity graph. In order to explore the structural information between the nodes, a transition probability matrix is defined, and random walks are performed, finally the potential relationship between the hedonic game clusters is discovered by analyzing the random walk trajectory of different nodes. Experimental results on multiple datasets demonstrate the effectiveness of the proposed method.

关 键 词：聚类集成 hedonic games 随机游走 概率轨迹 软子空间聚类 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]