4.7 Article

CNIM-GCN: Consensus Neighbor Interaction-based Multi-channel Graph Convolutional Networks

期刊

EXPERT SYSTEMS WITH APPLICATIONS
卷 226, 期 -, 页码 -

出版社

PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.eswa.2023.120178

关键词

Network representation learning; Deep learning; Graph convolutional networks; Node classification

向作者/读者索取更多资源

Node classification is important in network applications and has gained attention recently. CNIM-GCN is a novel method that preserves common information between feature space and topology space explicitly by introducing a consensus graph. Experimental results show that CNIM-GCN outperforms existing baselines.
Node classification plays a critical role in numerous network applications, and has attracted increasing attention in recent years. Existing state-of-the-art studies aim at maintaining common information between the topology graph and the feature graph in an implicit way, i.e., adopting a common convolution with parameter sharing strategy to preserve common information between the two graphs. Despite their effectiveness, these studies are still far from satisfactory due to the complex correlation information between the two spaces. To address this issue, we present a novel method named Consensus Neighbor Interaction-based Multi-channel Graph Convolutional Networks (CNIM-GCN). CNIM-GCN preserves the common information between the feature space and topology space in an explicit way by introducing a consensus graph for information propagation. A multi-channel graph convolutional networks is developed for effectively fusing information from different graphs. In addition, we further incorporate two types of consistency constraints, i.e., structural consistency constraint and reconstruction consistency constraint, to maintain the consistency between different channels. The former is leveraged to keep the consistency between different spaces at the structural relationship level, while the latter is used to preserve a consistency between the final node representation and the original node feature representation. We carry out extensive experiments on five real-world datasets, including ACM, BlogCatalog, CiteSeer, Flickr and UAI2010. Experimental results show that our proposed approach CNIM-GCN is superior to the state-of-the-art baselines.

作者

我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。

评论

主要评分

4.7
评分不足

次要评分

新颖性
-
重要性
-
科学严谨性
-
评价这篇论文

推荐

暂无数据
暂无数据