Abstract:Multi-view clustering can improve clustering performance by leveraging the complementary information from multiview data and has garnered growing interest. Despite significant advancements has been established in this field, most existing methods primarily concentrate on learning the latent representation, neglecting the latent graph that typically reveals relationships between different instances. To address this limitation, this paper presents a new framework of Deep Multi-View Clustering via View-specific Representation and Global Graph (DVGMVC) to address the aforementioned issue. Our method leverages the latent graph of each view to enhance representation quality and incorporates a global graph that combines discriminative information to enhance the performance of deep embedded clustering. Specifically, we employ autoencoders to learn embedded features and construct latent graphs from each view. Subsequently, a weighted two-layer plain residual network refines the latent feature using topological information derived from the latent graphs. As various views adhere to the same foundational cluster structure across multiple views, DVGMVC merges the embedded features of all views to create a unified feature. By leveraging the Graph Convolutional Network (GCN) module, which exploits information from both the latent graphs and the global graph generated from the global feature, a unified embedded feature is generated to capture comprehensive topological information among samples. This unified embedded feature is then employed to generate pseudo labels in a self-supervised manner, guiding unified embedding learning and clustering. In a self-supervised manner, such pseudo-label information guides unified embedding learning and clustering. In-depth experiments on public datasets has proven the state-of-the-art clustering effectiveness of our proposed method.

Deep Graph Reconstruction for Multi-View Clustering.

Deep Multi-View Clustering via View-Specific Representation and Global Graph

Deep multi-view graph clustering with incomplete views

Graph-Driven Deep Multi-View Clustering with Self-Paced Learning

Dynamic Weighted Graph Fusion for Deep Multi-View Clustering

Multi-view Clustering with Deep Matrix Factorization and Global Graph Refinement

Dual Fusion-Propagation Graph Neural Network for Multi-View Clustering

Deep Graph Regularized Non-Negative Matrix Factorization for Multi-View Clustering

Deep Multi-View Spectral Clustering Via Ensemble

Learning Latent Disentangled Embeddings and Graphs for Multi-View Clustering

Multi-view Projected Clustering with Graph Learning.

Dual-Optimized Adaptive Graph Reconstruction for Multi-View Graph Clustering

Consensus Representation-Driven Structured Graph Learning for Multi-View Clustering

GRAE - Graph Recurrent Autoencoder for Multi-view Graph Clustering.

Fast Multiple Graphs Learning for Multi-View Clustering

Effective and Efficient Graph Learning for Multi-view Clustering

Dual Label-Guided Graph Refinement for Multi-View Graph Clustering.

Multi-View Graph Attention Embedding for Clustering

Dropping Pathways Towards Deep Multi-View Graph Subspace Clustering Networks

Feature Weighted Multi-View Graph Clustering

Consistent Multiple Graph Embedding for Multi-View Clustering