Abstract
Recently, anchors-based multi-view clustering methods have been widely concerned for they can not only significantly reduce the time complexity but also have good interpretability. However, the time consumption of optimization and spectral embedding with Singular Value Decomposition (SVD) is expensive for large-scale multi-view clustering due to the large-scale consensus graph. This paper proposes to factorize the consensus graph to directly obtain the low-dimension consensus embedding matrix by optimizing the objective function. Specifically, the consensus graph is factorized into a transition matrix and a low-dimension embedding matrix. Among them, the transition matrix is used to prevent the clustering time consumption from increasing significantly with the number of anchors, and the low-dimension embedding matrix is used to mine the low-dimension consensus information of each view. The proposed method demonstrates its superiority by outperforming eight multi-view clustering algorithms on nine datasets, as evidenced by the clustering results.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call