Multi-view clustering via anchor graph matrix tri-factorization

Multi-view clustering shows great potential by integrating complementary information from multiple views. Among them, anchor graph-based methods are widely used because they can efficiently approximate global data structures and maintain good scalability on large-scale datasets. However, existing methods usually treat the anchor graph matrix as a whole and do not explore its internal structure in depth. To address this challenge, we propose a scalable and robust multi-view clustering framework called MVC-AGMTF (Multi-View Clustering via Anchor Graph Matrix Tri-Factorization). We first construct anchor graphs for each view to efficiently capture global data structure. The resulting anchor graph matrix is factorized into three low-rank components under orthogonality constraints, modeling the shared subspace structure more effectively. To reinforce consistency among different views and uncover shared structural patterns, we incorporate a Schatten-p norm regularization on the resulting tensor representation. A series of experiments conducted on standard benchmark datasets validate the effectiveness and scalability of our method, especially in scenarios involving large-scale data.