多样性约束和高阶信息挖掘的多视图聚类  

Multi-view clustering with diversity constraints and high-order information mining

作　　者：赵振廷 赵旭俊[1] Zhao Zhenting;Zhao Xujun(School of Computer Science&Technology,Taiyuan University of Science&Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学计算机科学与技术学院,太原030024

出　　处：《计算机应用研究》2024年第8期2309-2314,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61572343);山西省基础研究计划资助项目(202303021221142)。

摘　　要：在现有的多视图聚类研究中,大多数方法没有考虑多视图的多样性,也没有关注数据的高阶邻域信息,导致聚类结果不够准确,难以挖掘数据集的底层信息。为了解决这些问题,提出了基于多样性约束和高阶信息挖掘的多视图聚类算法(MVCDCHO)。首先设计了视图间多样性测量的方法,利用多样性的约束保留数据的交集特征,同时去除多视图的差异特征;然后提出了一种挖掘视图高阶信息的方法,要求多视图的交集特征接近混合相似图,以挖掘数据间相关性所没有关注到的高阶信息;最后将多视图的交集特征融合成共识图,通过谱聚类来获取聚类目标图;另外,设计了一种交替迭代的方法来迭代学习优化目标函数。实验结果表明,MVCDCHO在归一化互信息(NMI)、调整后的兰德指数(ARI)、聚类精度(ACC)多个聚类评价指标上表现出优异的性能。理论分析和实验研究验证了MVCDCHO中多视图多样性和高阶信息的关键作用,证明了MVCDCHO的优越性。In the current research on multi-view clustering,the majority of methods have not adequately considered the diversity of multiple views nor focused on the high-order neighborhood information of the data,which leds to clustering results that lack accuracy and struggle to uncover the underlying information in datasets.To address these issues,this paper proposed a multi-view clustering method based on diversity constraints and high-order information mining(MVCDCHO).Firstly,it designed a method for measuring diversity between views,utilizing diversity constraints to preserve the intersection features of the data while eliminating differing features across multiple views.Subsequently,it introduced a method for mining high-order information in views,requiring the intersection features of multiple views to approximate a mixed similarity graph,thereby extracting high-order information in data correlations that has been overlooked.Finally,it fused the intersection features of multiple views into a consensus graph and employ spectral clustering to obtain the clustering target graph.Additionally,it designed an alternating iterative method,iteratively learning to optimize the objective function.The experimental results show that MVCDCHO has excellent performance on the normalized mutual information(NMI),the adjusted Rand index(ARI),and the clustering accuracy(ACC).Theoretical analysis and experimental study underscore the crucial role of multi-view diversity and high-order information in the MVCDCHO algorithm,providing evidence for its superiority.

关 键 词：多视图聚类 多样性 一致性 高阶信息 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]