Node-Feature Convolution for Graph Convolutional Networks

• A novel strategy to construct a fixed-size feature map via neighbor selection and ordering. new node-feature convolution (NFC) layer for graph convolutional network (GCN). Insightful studies on varying aggregators, neighborhood size, model depth. Demonstration of the efficacy NFC-based GCNs benchmark datasets. Graph (GCN) is an effective neural representation learning. However, standard GCN suffers from three main limitations: (1) most real-world graphs have no regular connectivity node degrees can range one hundreds or thousands, (2) neighboring nodes are aggregated with fixed weights, (3) features within vector considered equally important. Several extensions been proposed tackle limitations respectively. This paper focuses tackling all limitations. Specifically, we propose GCN. The NFC first constructs using selected ordered number neighbors. It then performs operation this learn representation. In way, usefulness both individual neighborhood. Experiments datasets show that NFC-GCN consistently outperforms state-of-the-art methods in classification.