Large-Scale Multi-View Multiple Clustering

Abstract

Multi-view multiple clustering generates multiple clustering results to uncover diverse information in multi-view data, but existing solutions for this task are inefficient when processing large-scale data, and it is difficult to achieve high-quality and diverse clustering simultaneously. Therefore, we propose the Large-Scale Multi-View Multiple Clustering (LSMVMC) algorithm. Inspired by anchor graph techniques, we construct a relationship matrix using representative anchors in the neighborhood structure of the original samples, improving clustering speed without compromising quality. Deep matrix factorization decomposes the anchor graph into multiple subspaces through a multi-layer decomposition matrix. By assigning different weights to views, comprehensive consideration of view-specific information enables the generation of high-quality clusters in each subspace. We enhance cluster diversity by employing orthogonal subspaces to comprehensively consider redundancy. Experimental results on benchmark datasets demonstrate that our solution is generally more efficient and performs better than other large-scale multi-view data processing methods.