双共识多视角谱聚类  

Co-consensus Multi-view Spectral Clustering

作　　者：陈曙 朱正东 杨祖元[1] 李珍妮 Chen Shu;Zhu Zheng-dong;Yang Zu-yuan;Li Zhen-ni(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)

机构地区：[1]广东工业大学自动化学院,广东广州510006

出　　处：《广东工业大学学报》2024年第4期98-105,共8页Journal of Guangdong University of Technology

基　　金：国家自然科学基金资助项目(62273106)。

摘　　要：多视角学习因其能融合各视角信息而受到广泛关注。针对多视角数据融合问题,本文提出了一种双共识多视角谱聚类方法,在谱聚类模型中添加两种共识约束,利用不同视角谱嵌入矩阵的特征关系和相似关系,增强多视角之间的一致性。同时,该方法在优化过程中能够获得相应共识变量的闭式解,进一步提升了聚类性能。实验在3个真实世界数据集中测试了该方法的收敛性及对参数的敏感性和聚类效果。实验结果表明,与现有的方法相比,本文的方法在多个性能指标上都有更好的表现,在聚类精度上最高提升超过10%。使用双共识方法可以有效提高多视角谱聚类的性能。Multi-view learning has attracted wide attention because of the ability to integrate information from different views.For the issue of multi-view data fusion,a co-consensus multi-view spectral clustering method is proposed.The method adds two consensus constraints in the model of spectral clustering to utilize the feature relationship and the similarity relationship of different views’spectral embedding matrices which enhances the consistency of multi views.Simultaneously,this method obtains closed solution of the consensus variables in the optimization process,which further improves the clustering performance.The experiment tests the convergence,parameter sensitivity and clustering performance of the proposed method in three real-world datasets.The experiment results show that this method has the best performance in multiple performance metrics compared with the existing methods,and the maximum improvement in clustering accuracy is more than 10%.The experiment proves the co-consensus method effectively improves the performance of multi-view spectral clustering algorithm.

关 键 词：多视角学习 共识 谱聚类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]