<p>Multi-view clustering (MVC) aims to exploit complementary information from heterogeneous views to uncover consistent cluster structures. Despite recent progress, existing contrastive learning–based MVC methods still suffer from two fundamental limitations: global semantic misalignment caused by independently optimized view-specific clusters, and performance degradation induced by noisy and inconsistent pseudo-labels across views. To address these issues, we propose Multi-View Clustering via Hybrid Matrix Factorization and Label Correction (MFLC), a unified framework that explicitly aligns cross-view cluster semantics while dynamically correcting unreliable pseudo-labels. MFLC formulates cross-view semantic coordination as a matrix factorization problem, in which view-specific features are decomposed into centroid-oriented latent factors using a FunkSVD-based scheme, enabling global semantic alignment without sacrificing view-specific characteristics. In addition, a dual-granularity contrastive learning strategy is employed to jointly enforce instance-level feature consistency and cluster-level semantic agreement, while a Hungarian-based label correction mechanism establishes permutation-invariant correspondences between factorization-derived cluster assignments and contrastive predictions to further suppress pseudo-label noise. Extensive experiments on five benchmark datasets demonstrate that MFLC consistently outperforms state-of-the-art methods in terms of clustering accuracy, normalized mutual information, and robustness to noisy views.</p>