A probability-driven structure-aware algorithm for influence maximization under independent cascade model

Influence maximization (IM) is the problem of finding a set of nodes that can achieve the maximal influence spreads into the network, which faces two significant but intractable issues in latest studies: (i) Curse of scales: with the increase of the network scale, traditional methods cost extensive times in guaranteeing accuracy, which re-evaluate influence spread of every node in network, leading to significant computational overhead; (ii) Generalization issue: with more and more studies on various networks and propagation parameters, it is difficult to find a universally appropriate algorithm that performs well in each topology. In this paper, we propose a novel probability-driven structure-aware (PDSA) algorithm, which begins by cutting/updating network according to the edge activation probability parameters of the IC model, and then uses a graph traversal algorithm (e.g., breadth first search algorithm) to evaluate the influence spread scores of each node. Meanwhile, we adopt a kind of centrality-based independent cascade (CIC) model to approximate a more realistic propagation scenario. Through extensive experiments with six real-world/synthetic networks and six CIC/IC models, we demonstrate that PDSA achieves great performance over state-of-the-art algorithms in terms of effect and efficiency. Even facing various complex topologies and propagation parameters, PDSA exhibits excellent robustness in solving IM problems.