Self-consistent Contrastive Attributed Graph Clustering with Pseudo-label Prompt

Attributed graph clustering, which learns node representation from attribute and topological for is a fundamental challenging task multimedia network-structured data analysis. Recently, contrastive learning (GCL)-based methods have obtained impressive clustering performance on this task. Nevertheless, there still remain some limitations to be solved: 1) most existing fail consider the self-consistency between latent representations cluster structures; 2) require post-processing operation get labels. Such two-step scheme results in models that cannot handle newly generated data, i.e. , out-of-sample (OOS) nodes. To address these issues unified framework, S elf-consistent xmlns:xlink="http://www.w3.org/1999/xlink">C ontrastive xmlns:xlink="http://www.w3.org/1999/xlink">A ttributed xmlns:xlink="http://www.w3.org/1999/xlink">G raph lustering (SCAGC) network with pseudo-label prompt proposed article. In SCAGC, by labels information, self-consistent loss, aims maximize consistencies of intra-cluster while minimizing inter-cluster representations, designed learning. Meanwhile, module built directly output contrasting different clusters. Thus, OOS nodes, SCAGC can calculate their Extensive experimental seven benchmark datasets shown consistently outperforms 16 competitive methods. The source code could accessed at https://github.com/xdweixia/SCAGC .