A multifactorial evolutionary algorithm to detect stably influential seeds from competitive networks under multiple damage scenarios

Abstract

The complex network has garnered significant attention over the past few decades, with optimization and information extraction problems show significance in practical applications. The competitive influence maximization, along with its robustness, has emerged as a recent focal point, which is aimed at identifying seeds with robust and influential capabilities across multiple propagative groups within a specific network. Existing studies indicate the damage percentage of link-based failures can be pre-defined, and solving the problem in a single-objective manner. However, it has been demonstrated that multiple damage scenarios are prevalent, and the corresponding search processes may yield the synergy. Therefore, the correlation between the optimization directed at different damage scenarios of link-based attacks is analyzed first, which has shown non-conflict relation. Consequently, the multitasking optimization paradigm is thus introduced to modeling the related robust influence maximization problem. A numerical metric is also designed to reflect the significance of links on competitive networks. Equipped with this metric, a multifactorial evolutionary algorithm has been developed to tackle the seed determination problem under multiple damage scenarios, termed MFEA-RCIM_MD. The involved operators consider diverse information from both genetic and fitness domains, and a multi-phase transfer operation is included to leverage knowledge across different tasks. Experiments on synthetic and real-world networks demonstrate the remarkable performance of the algorithm over existing single-objective and multitasking approaches. With enhanced efficiency, multiple candidates are provided for decision-makers to address diffusive challenges in practical systems.