Graph embedding clustering: Graph attention auto-encoder with cluster-specificity distribution

Towards exploring the topological structure of data, numerous graph embedding clustering methods have been developed in recent years, none them takes into account cluster-specificity distribution nodes representations, resulting suboptimal performance. Moreover, most existing execute representations learning and two separated steps, which increases instability its original Additionally, rare simultaneously node attributes reconstruction account, degrading capability learning. In this work, we integrate a unified framework, propose new deep attention auto-encoder for that attempts to learn more favorable by leveraging self-attention mechanism reconstruction. Meanwhile, constraint, is measured ?1,2-norm, employed make within same cluster end up with common dimension space while different clusters distributions intrinsic dimensions. Extensive experiment results reveal our proposed method superior several state-of-the-art terms