Exploring an edge convolution and normalization based approach for link prediction in complex networks

Link prediction in complex networks is to discover hidden or to-be-generated links between network nodes. Most of the mainstream graph neural (GNN) based link methods mainly focus on representation learning nodes, and are prone over-smoothing problem. This paper dedicates links, designs an edge convolution operation so as realize learning. Besides, we propose normalization strategy for learned representation, purpose alleviating problem model, when constructing EdgeConvNorm with stacking manipulations. Lastly, employ a binary classifier sigmod Hadamard product two nodes parsed from final representation. The can also be employed baseline, extensive experiments real-world benchmark validate that not only alleviates problem, but has advantages over representative baselines.