Multi-view unsupervised feature selection with consensus partition and diverse graph

Abstract

Multi-view unsupervised feature selection has gained significant attention in effectively reducing the dimensionality of unlabeled data collected from multiple sources. Many existing methods integrate the tasks of graph learning and feature selection to select informative features. While these methods have demonstrated promising results, they usually construct a consensus graph by merging multiple graphs or consider the consistent graph of all views. However, due to the view heterogeneity, it becomes difficult to identify a shared similarity structure. On the other hand, the clustering outcome remains consistent across all views. In light of this, we propose generating multiple graphs that are as mutually exclusive as possible to enhance the complementarity between views. Additionally, our method bridges graph learning and consensus clustering to leverage the indicator consistency. We also present an effective algorithm to optimize the objective function. Finally, extensive experiments on six benchmark datasets demonstrate that our method outperforms state-of-the-art methods. The code is available at https://github.com/HdTgon/2023-INS-CDMvFS.