基于双锚点图的多视图模糊聚类

打开文本图片集

中图分类号：TP391.4 文献标志码：A 文章编号：1673-2340（2025）03-0064-11

Abstract：Inrecent years，with therapiddevelopmentof multi-view learning，howtoefectivelyintegrate information from different views for clustering analysis has becomean important research topicin both academia and industry， driving the emergence of numerous eficient methods.However，，curent methods stillface three major chalenges. First，suboptimal anchor graphs often result from the inherent uncertaintyand low discriminabilityof real-world data. Second，prevalentapproaches primarily focusoncommon information between views，overlooking valuableview-specific information.Third，effectively leveraging the learned anchor graph to improve clustering remainsunder-explored. Toovercome thesechalenges，this paper proposes a novel dual anchor graph fuzzy clustering framework.To address thefirsttwo challnges，we designamatrix factorization-based dual anchor graph learning framework.This framework extracts discriminative hidden representations fromeach view and subsequently derives bothcommon and specificanchor graphs.For the third chalenge，we develop an anchor graph fuzzy clustering method with acooperative learning mechanism.This method constructs a dual anchor graph-driven fuzzy membership structure preservation mechanism to enhance clustering quality.Additionally，we introducenegative Shannon entropy toachieve adaptive view weighting.Extensiveexperiments on multiplebenchmark datasetsvalidatethe effectiveness of the proposed DAG_FC method.Theresultsshow.that DAG_FC outperformscompeting methodsonmost metricsand datasets， achievingNMI improvementsof approximately 30% and 20% over comparative methods on the Yale dataset.Moreover，theexperimentsalso confirm that anchor graph-based clustering methods generally perform beter than traditional subspac-based clustering methods.By incorporating hidden representation extraction techniques and designing specialized clustering algorithms，this paper further enhances the clustering performance of the proposed method.

Key words： multi-view data; dual anchor graph learning; common information;specific information; fuzzy clustering

随着数字科学技术的发展，可采集数据的多样性也随之显著增加。（剩余18790字）