Abstract: The effectiveness and applicability of multi-view clustering algorithms for multi-view data processing required further improvement. Most existing multi-view clustering algorithms were hindered by incomplete fusion mechanisms, insufficient exploration of multi-view collaborative relationships, and unstable robustness. These limitations led to lower consistency in clustering outcomes and reduced stability when handling noise and redundant information. To address these challenges, a novel Multi-View Clustering Algorithm Based on Bipartite Graph and Consistency Graph Learning (BGC-MVC) was developed to enhance clustering consistency and complementarity by effectively integrating information from multiple views. A bipartite graph was first constructed to capture the neighborhood relationships between different views. Subsequently, a consensus graph was learned to strengthen inter-view similarity. The embeddings of the original multi-view data were integrated into a unified framework, combining graph learning with the clustering process to improve overall clustering performance. Experimental results demonstrated that significant improvements in evaluation metrics were achieved under convergence conditions. On the MSRC_v1 dataset, the F-score increased by 19.84 percentage points compared to the second best LMVSC algorithm. Furthermore, BGC-MVC exhibited superior robustness and accuracy.

Key words: multi-view clustering, bipartite graph, consensus graph, graph fusion, embedding learning

摘要： 当前多视图聚类算法在多视图数据处理的有效性与适用性方面还有进一步提升的空间，。目前大多数多视图聚类算法存在融合机制不够完善、对多视图协同关系挖掘不足以及鲁棒性不够稳定等问题，导致聚类结果一致性偏低，且在噪声和冗余信息下性能不够稳健。针对上述问题，提出一种基于二部图和一致图学习的多视图聚类算法（BGC-MVC），该算法旨在通过融合各视图信息，从而提升聚类的一致性和互补性。首先，通过构造二部图以捕获不同视图之间的邻域关系；其次，学习一致性图以强化视图间的相似性。BGC-MVC算法将原始多视图数据的嵌入整合进一个统一的框架中，将图学习与聚类过程结合起来，从而提高聚类的整体效果。实验结果表明，BGC-MVC算法在满足收敛性的条件下各评价指标的值有明显的提升，在MSRC_v1数据集上的F-score比次优的LMVSC算法提高了19.84个百分点，并且BGC-MVC算法表现出更强的鲁棒性与准确性。

关键词: 多视图聚类, 二部图, 一致图, 图融合, 嵌入学习