基于聚类结构和局部相似性的多视图隐空间聚类  

Multi-view latent subspace clustering with cluster structure and local similarity

作　　者：宋菲[1] Song Fei(School of Information Technology,Jiangsu Open University,Nanjing 210017,China)

机构地区：[1]江苏开放大学信息工程学院,南京210017

出　　处：《计算机应用研究》2023年第9期2650-2656,共7页Application Research of Computers

基　　金：国家自然科学基金青年科学基金资助项目(62206114)。

摘　　要：随着数据获取方式的多样化发展,针对多视图领域的算法研究变得越来越重要,但大多数方法仅通过自表示属性或局部结构获取样本间的相似性关系,在此过程中忽略了整体样本的聚类结构和原始空间的噪声的影响,使得聚类结果存在较大误差。为解决此问题,提出了一种基于聚类结构和局部相似性的多视图隐空间聚类方法(multi-view latent subspace clustering with cluster structure and local similarity,MLC2L),通过隐表示融合不同视图上的共享信息并抑制噪声的存在。此外,通过探索隐空间内样本间的局部相似性关系和整体的聚类结构促进样本达到同类聚合、异类远离的目的;最后引入一个交替方向迭代优化算法来快速求解目标函数。实验结果显示,在六个真实数据集的实验中,MLC2L在MSRC-v1、UCI以及100Leaves上的五个评价指标均为最优,在3Sources、WebKB和Prokaryotic等数据集上的五个指标有四个最优,大量的实验分析也证明了融合局部结构和整体聚类结构的MLC2L在多视图聚类任务上的有效性。In recent years,with the diversification of data acquisition,multi-view learning has become more and more important.Most multi-view clustering methods obtain the similarity between samples through self-representation or local structures.However,these methods don’t consider the influence of noise and the clustering structure of the overall sample,which may lead to a large error in the clustering study.To address this issue,this paper proposed a multi-view latent subspace clustering with cluster structure and local similarity(MLC2L),which combined shared information on different views and suppresses the presence of possible noise through latent representations.Besides,it simultaneously explored the clustering structure and local similarity in the latent space,so the samples could be promoted to achieve the purpose of homogeneous aggregation and heterogeneous separation.Further,this paper introduced an alternate direction iterative optimization algorithm to quickly solve the objective function.The experimental results in six real datasets show that the proposed method is optimal for five evaluation metrics on MSRC-v1,UCI,and 100Leaves,and four out of five metrics on 3Sources,WebKB,and Prokaryotic datasets.Extensive experimental results demonstrate the effectiveness of the MLC2L,which combines local structure and overall clustering structure,in multi-view clustering tasks.

关 键 词：多视图聚类 隐空间 聚类结构 局部相似性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]