Figure 1

A GNN framework based on edge convolution, designed for link prediction in complex networks and named EdgeConvHiF, combines both high- and low-frequency information. It should be noted that this article only uses the representation aggregation and transformation process of node \(v_1\) to illustrate how to fuse the high- and low-frequency graph information of nodes for node representation learning, and this process corresponds to Eq. 11, as illustrated in red box. Other nodes follow the same aggregation and transformation process.