Abstract

We extend the graph convolutional network method for deep learning on graphdata to higher order in terms of neighboring nodes. In order to constructrepresentations for a node in a graph, in addition to the features of the nodeand its immediate neighboring nodes, we also include more distant nodes in thecalculations. In experimenting with a number of publicly available citationgraph datasets, we show that this higher order neighbor visiting pays off byoutperforming the original model especially when we have a limited number ofavailable labeled data points for the training of the model.

 

Quick Read (beta)