Multi-view graph clustering with Dually Enhanced Tensor Rank Minimization and Diverse Separation of Inconsistent...

Multi-view graph clustering is a powerful machine-learning technique for data analysis. However, most of the previous methods still suffer from several limitations. First, most methods overlook the potential inconsistent information in multiple views as well as their diversity across views. Second, they usually lack the ability to capture the high-order correlation between and within views simultaneously. Third, most tensor-based methods tend to use the Tensor Nuclear Norm with the same penalty for different singular values to approximate the rank of the tensor. To tackle these limitations, this paper proposes a new multi-view graph clustering method, which takes into account the diversified separation of inconsistent information in the view and introduces the Enhanced Tensor Rank minimization to consider the high-order correlation within and between views at the same time. To optimize the model, we design an effective minimization method. Remarkably, the experimental results on multiple datasets verify the superiority of the proposed method.