Link prediction in multiplex social networks based on information transmission graph neural networks

Social networks are omnipresent graph-structured data where entities and interactions between them can be modeled as nodes and edges, respectively. In recent years, graph neural networks (GNNs) have evolved as a mainstream and powerful model for handling graph data. Link prediction, which can foretell missing or potential links, is a crucial task in social network analysis. However, most existing methods focus on monolayer networks and usually ignore that multiplex networks have richer information, which often results in poor performance on link prediction. Therefore, this paper proposes an information transmission-based GNN (IT-GNN) model for link prediction in multiplex social networks. IT-GNN first learns the representation of each node in a multiplex network from itself and its neighborhood. Specifically, the neighborhood information of a node is learned from its 1-, 2- and 3-hop neighbors in combination with the three new ways of information transmission. Then, the embedding of a node pair is obtained based on these representations. In this regard, the final representations of potential links in the target layer, which are used to determine the existence of potential links, are generated by fusing their embeddings from all layers. Extensive experiments on seven real-world networks demonstrate the superior performance of the proposed model.