Extending influence maximization by optimizing the network topology

Influence maximization has recently obtained significant attention in advertising and rumor control on a social network. However, most existing works on influence maximization focus on mining seed nodes with high influence, while the timeliness of information diffusion in dynamic social networks has not been fully explored, resulting in the diffusion effect is not up to the expected level. To fill this gap, we investigate how to maximize the influence spread range of seed set A by adding k latent edges to dynamic network G which is called Dynamic Edge Addition (DEA). For this purpose, we design an Activation Probability-aware (AP) framework for adding edges to the network. Specifically, AP contains three modules: graph generation, influence estimation of nodes, and edge addition. The module of graph generation is used to obtain graph structure. In the module of influence estimation, we employ a baseline Influence Maximization via Martingales (IMM) algorithm and then improve it to make the seed more effective by setting a limit to the traversal depth of sampling. To capture the dynamic temporal changes of the network as soon as possible, we dynamically update the influence of nodes according to the propagation probability of the path. In the module of edge addition, we propose a novel strategy of edge addition by finding the shortest path length between the potential users and seed users, and realize the strategy in linear time. Extensive experiments on real-world datasets are conducted and the results show the superiority of the AP over other baselines.