Anchor-based clustering methods have emerged as an effective paradigm for improving clustering efficiency, particularly in the context of large-scale data analysis. By introducing a compact set of representative samples, these methods significantly reduce computational complexity while preserving essential structural information within the original data. However, in multi-view scenarios, heterogeneous data distributions often lead to anchor misalignment, where anchors from different views fail to establish accurate correspondence. This Anchor Unalignment Problem (AUP) causes an incorrect graph fusion and disrupts the global structural consistency across views, ultimately degrading clustering performance. To this end, we propose a novel framework, named Threefold Consensus-Driven Anchor Alignment for Efficient Multi-View Clustering (TCAA), which jointly enforces representation consistency, spectral consistency, and discrete clustering consistency across views under an adaptive weighting scheme. Specifically, permutation matrices are employed to align anchor graphs across views under the guidance of representation similarity, while spectral structural signatures capture the global roles of anchors to ensure robust alignment. Furthermore, discrete clustering consistency directly couples the consensus anchor graph with cluster assignments, eliminating the information loss of two-stage methods. Extensive experiments on real-world datasets demonstrate that TCAA outperforms state-of-the-art multi-view clustering methods.
原文链接
分享
引用
