Attribute Multiplex Network Graph Clustering: Joint Contrastive And High-Order Proximity

In this paper, we propose an improved subspace clustering algorithm, with the aim to improve accuracy and smoothness. To enhance the discriminative power of the coefficient matrix and broaden its applicability to various types of data, the algorithm incorporates contrastive loss and higher-order proximity, aiming to explore distinguishable subspace structure between data points. We first use graph convolution to design a low-pass filter to generate a smooth representation of the attribute view. Next, to focus on the multi-view consistency and the cross-view diversity, we introduce the contrastive loss function and high-order neighbor relations as regularizers. Once the optimal representation coefficient matrix Z has been determined, Z is used to construct the affine matrix for spectral clustering. After experimental verification on multiplex graphs datasets, our method shows higher accuracy and smoothness than current algorithms.