A Multi-Factorial Evolutionary Algorithm With Asynchronous Optimization Processes for Solving the Robust Influence Maximization Problem.

The complex network has attracted increasing attention and shown effectiveness in modeling multifarious systems. Focusing on selecting members with good spreading ability, the influence maximization problem is of great significance in network-based information diffusion tasks. Plenty of attention has been paid to simulating the diffusion process and choosing influential seeds. However, errors and attacks typically threaten the normal function of networked systems, and few studies have considered the influence maximization problem under structural failures. Therefore, a quantitative measure with a changeable parameter is first developed in this paper to tackle the unpredictable destruction percentage on networks. Further, limitations on the existing methods are shown experimentally. To address these limitations, the evolutionary multitasking paradigm is employed, and several problem-specific operators are developed. On top of these developments, a multi-factorial evolutionary algorithm is devised to find seeds with robust influence ability, termed MFEARIM, where the genetic information for both myopia and holistic areas is considered to improve the search ability. Additionally, an asynchronous strategy is designed to efficiently tackle tasks with distinct costs, and the convergence of the search process can thus be accelerated. Experiments on several synthetic and real-world networks validate the competitive performance of MFEARIM over the existing methods. [ABSTRACT FROM AUTHOR]