Fast Multi-view Clustering with Sparse Matrix and Improved Normalized Cut

Abstract

The multi-view clustering algorithm is a novel approach to explore the inherent clustering structure among data. However, most existing methods suffer from noise issues when constructing similarity graphs and may lose important information during the clustering, leading to lower accuracy. Moreover, iterative optimization approaches often used by these algorithms can be memory-overflowing and time-consuming. To address these limitations, a fast multi-view clustering algorithm with sparse matrix and improved normalized cut (SINFMC) is proposed. It first constructs similarity graphs for all views and integrates them to form a consensus graph matrix. Then, the [l1]-norm constraint is applied to the consensus graph matrix to obtain a sparse matrix, which helps to denoise the data and speed up computations. Finally, an improved normalized spectral clustering algorithm is used to cluster the sparse consensus graph and obtain a cluster indicator matrix. This matrix provides clustering results directly and avoids information loss and bias. Unlike other methods, the proposed algorithm does not require iterative optimization and simplifies the computation process through sparse matrix representation, reducing time and space complexity. Experimental results on both artificial and real-world datasets demonstrate that the proposed algorithm outperforms the compared algorithms in terms of quality and efficiency.

Keywords

• multi-view clustering; sparse matrix; normalized cuts; soft threshold; graph fusion