Graph Convolutional Network for Recommendation with Low-pass Collaborative Filters

\\textbf{G}raph \\textbf{C}onvolutional \\textbf{N}etwork (\\textbf{GCN}) is\nwidely used in graph data learning tasks such as recommendation. However, when\nfacing a large graph, the graph convolution is very computationally expensive\nthus is simplified in all existing GCNs, yet is seriously impaired due to the\noversimplification. To address this gap, we leverage the \\textit{original graph\nconvolution} in GCN and propose a \\textbf{L}ow-pass \\textbf{C}ollaborative\n\\textbf{F}ilter (\\textbf{LCF}) to make it applicable to the large graph. LCF is\ndesigned to remove the noise caused by exposure and quantization in the\nobserved data, and it also reduces the complexity of graph convolution in an\nunscathed way. Experiments show that LCF improves the effectiveness and\nefficiency of graph convolution and our GCN outperforms existing GCNs\nsignificantly. Codes are available on \\url{https://github.com/Wenhui-Yu/LCFN}.\n