Hybrid Low-Order and Higher-Order Graph Convolutional Networks

Figure 2

HLHG mode. The graph convolutional network layer of the HLHG model consists of two convolutional layers and information fusion pooling. The input parameters are from the first-order to the n-th order neighborhoods. When n = 1, the model degenerates into a classical graph convolution GCN model. When the neighborhood order is n = 2, it is called the HLHG-2 model, and its input parameters are the 1st order neighborhood and the 2nd order neighborhood. When the neighborhood order is n = 3, it is called the HLHG-3 model, and its input parameters are the 1st order neighborhood, the 2nd order neighborhood, and the 3rd order neighborhood.