Link Prediction with Mining and Filtering Information in Multiplex Networks

A special kind of complex networks known as multiplex networks occurs when a single collection of entities engages in a variety of interactions. Moreover, in a multiplex network, the topologies of other levels may have an impact on how links form in one layer. At present, a lot of solutions to the link prediction problem have been developed. However, few of them have tackled link prediction in multiplex networks using graph neural networks. In this paper, we propose a new approach to conduct link prediction in multiplex networks using graph neural networks. The proposed approach first combines all layers of a network to form an information aggregation graph, and then reconstruct an information filter graph by leveraging the similarities of nodes. Next, for the target node pair, we extract its enclosing subgraphs from the target layer, auxiliary layers, and information filtering graph, and then predict the connection of the target node pair by inputting the subgraphs with feature matrices into a multilayer perceptron. The proposed solution offers better performance for link prediction in multiplex networks, as shown by the experimental results on six real networks.