BiGCN: A bi-directional low-pass filtering graph neural network

Graph convolutional networks (GCNs) have achieved great success on graph-structured data. Many GCNs can be considered low-pass filters for graph signals. In this paper, we propose a more powerful GCN, named BiGCN, that extends to bidirectional filtering. Specifically, consider the original structure information and latent correlation between features. Thus BiGCN filter signals along with both feature-connection graph. Compared most existing GCNs, is robust has capacities feature denoising. We perform node classification link prediction in citation co-purchase three settings: Noise-Rate, Noise-Level, Structure-Mistakes. Extensive experimental results demonstrate our model outperforms state-of-the-art neural clean artificially noisy