A Novel Centrality Cascading Based Edge Parameter Evaluation Method for Robust Influence Maximization

The research of social influence is an important topic in online social network analysis. Influence maximization is the problem of finding k nodes that maximize the influence spread in a specific social network. Robust influence maximization is a novel topic that focuses on the uncertainty factors among the influence propagation models and algorithms. It aims to find a seed set with a definite size that has robust performance with different influence functions under various uncertainty factors. In this paper, we propose a centrality-based edge activation probability evaluation method in the independent cascade model. We consider four different types of centrality measurement methods and add a modification coefficient to evaluate the edge probability. We also propose two algorithms, called NewDiscount and GreedyCIC, by incorporating the edge probability space into previous algorithms. With extensive experiments on various real online social network data sets, we find that our PageRank-based greedy algorithm has the best influence spreads and lowest running times, compared with other algorithms, on some large data sets. The experiment for evaluating the robustness performance shows that all algorithms have optimal robustness performance when the modification coefficient is set to 0.01 under the independent cascade model. This result suggests some further research directions under this model.