The matrix factorization approaches have been widely applied for multi-view clustering since they can effectively explore complementary information contained in the multi-view data. However, some prior knowledge hidden in multi-view data cannot be fully exploited in existing matrix factorization based multi-view learning approaches. In this paper, we present a robust dual-graph regularized deep matrix factorization (RDDMF) approach for multi-view clustering. Specifically, it integrates the dual-graph regularizers and the sparse constraints into the deep matrix factorization framework. Therefore, the proposed RDDMF approach discovers the geometric structures of both the data and the feature space by adding the dual graph regularization term into deep matrix factorization in each layer. Meanwhile, the sparse constraints are imposed on the coefficient matrix of each layer to improve the robustness of our model. Besides, we design an efficient optimization strategy of the proposed model and give its convergence rate. Numerous experiments on four well-known datasets show our proposed RDDMF approach is superior to several state-of-the-art approaches in multi-view clustering.

矩阵分解方法已广泛应用于多视图聚类，因为它们可以有效地探索多视图数据中包含的互补信息。然而，在现有的基于矩阵分解的多视图学习方法中，无法充分利用隐藏在多视图数据中的一些先验知识。在本文中，我们提出了一种用于多视图聚类的鲁棒双图正则化深度矩阵分解 (RDDMF) 方法。具体来说，它将双图正则化器和稀疏约束集成到深度矩阵分解框架中。因此，所提出的 RDDMF 方法通过将对偶图正则化项添加到每一层的深度矩阵分解中来发现数据和特征空间的几何结构。同时，对每一层的系数矩阵施加稀疏约束，以提高我们模型的鲁棒性。此外，我们设计了所提出模型的有效优化策略并给出了其收敛速度。对四个著名数据集的大量实验表明，我们提出的 RDDMF 方法优于多视图聚类中的几种最先进的方法。