Enhancing the robustness of influential seeds towards structural failures on competitive networks via a Memetic algorithm

Abstract

Networked systems present the correlative relation between entities, and the related knowledge excavation problem has been greatly emphasized in the last decade. The determination of influential nodes is modeled as the influence maximization problem, and the robustness issue has also been studied. Meanwhile, a specific network may maintain multiple diffusive groups, and the robustness of seeds under this environment is important to solve propagation tasks. Rational diffusion models have been developed in related studies, and valid algorithms are devised to find effective seeds. However, existing studies only consider the node-based attacks, and omit the impact of link removals on the robustness of seeds under competitive environments. As an indispensable component of networks, links show significance in sustaining the functionality, the analysis on link-based attacks is thus necessary. Aiming at addressing this deficiency, an importance metric is designed to evaluate the centrality of links in the diffusion process on competitive networks. Equipped with this metric, a numerical measure is developed to evaluate the performance of a given seed set. Further, the structural information and genetic information is considered to guide the search process. Several problem-directed operators are given to combine valuable knowledge from both myopic and holistic solution spaces. A Memetic algorithm is constructed to tackle the robust influence maximization problem on competitive networks under link-based attacks, termed MA-RCIM-L. The algorithm is tested on both synthetic and real-world networks, whose competitiveness against existing approaches is demonstrated.