eduzhai > Applied Sciences > Engineering >

AM-GCN Adaptive Multi-channel Graph Convolutional Networks

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 11 pages

Abstract: Graph Convolutional Networks (GCNs) have gained great popularity in tacklingvarious analytics tasks on graph and network data. However, some recent studiesraise concerns about whether GCNs can optimally integrate node features andtopological structures in a complex graph with rich information. In this paper,we first present an experimental investigation. Surprisingly, our experimentalresults clearly show that the capability of the state-of-the-art GCNs in fusingnode features and topological structures is distant from optimal or evensatisfactory. The weakness may severely hinder the capability of GCNs in someclassification tasks, since GCNs may not be able to adaptively learn some deepcorrelation information between topological structures and node features. Canwe remedy the weakness and design a new type of GCNs that can retain theadvantages of the state-of-the-art GCNs and, at the same time, enhance thecapability of fusing topological structures and node features substantially? Wetackle the challenge and propose an adaptive multi-channel graph convolutionalnetworks for semi-supervised classification (AM-GCN). The central idea is thatwe extract the specific and common embeddings from node features, topologicalstructures, and their combinations simultaneously, and use the attentionmechanism to learn adaptive importance weights of the embeddings. Our extensiveexperiments on benchmark data sets clearly show that AM-GCN extracts the mostcorrelated information from both node features and topological structuressubstantially, and improves the classification accuracy with a clear margin.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×