Self-Supervised Graph Attention Networks for Deep Weighted Multi-View Clustering

As one of the most important research topics in unsupervised learning field, Multi-View Clustering (MVC) has been widely studied past decade and numerous MVC methods have developed. Among these methods, recently emerged Graph Neural Networks (GNN) shine a light on modeling both topological structure node attributes form graphs, to guide unified embedding clustering. However, effectiveness existing GNN-based is still limited due insufficient consideration utilizing self-supervised information graph information, which can be reflected from following two aspects: 1) models merely use feature fail realize that such also applied sample weighting; 2) usage generally aggregation models, yet it provides valuable evidence detecting noisy samples. To this end, paper we propose Self-Supervised Attention for Deep Weighted (SGDMC), promotes performance deep by making full information. Specifically, novel attention-allocating approach considers similarity developed comprehensively evaluate relevance among different nodes. Meanwhile, alleviate negative impact caused samples discrepancy cluster structures, further design sample-weighting strategy based attention as well between global pseudo-labels local assignment. Experimental results multiple real-world datasets demonstrate our method over approaches.