Fig. 2.

Illustration of the graph convolution process. A graph 𝓖 can be represented by an adjacency matrix 𝓐 and a feature matrix 𝓧. The GCN takes the two matrices as input to conduct graph convolution. We used the red node as an example to show the convolution process. The neighbors with the same order have the same color in graph 𝓖. The colors of features are the same as the corresponding nodes. For the l^th layer, the red edges of the input adjacency matrix 𝓐 indicate the neighbors that participate in the convolution process and the features of these activate neighbors are non-transparent.