Finding robust influential seeds from networked systems against structural failures using a niching memetic algorithm

Abstract

There has been a growing interest in analyzing the dynamics and characteristics of networked systems, and the influence maximization problem is a recent hotspot to determine influential seeds from nodal members. Widespread applicable values can be found in marketing and information digging tasks, which boost more attention on related studies. Several diffusion models and seed determination techniques have been developed. Meanwhile, networked systems are threatened by sabotages and structural perturbances, and recent literature indicated that the robustness of the diffusion process is significant in daily applications, which can be modeled as the robust influence maximization problem. However, existing studies have not touched upon the node-based attack and its impact on the performance of seeds; corresponding robust seeds determination strategies are also lacked. For solving the robust influence maximization problem considering node-based attacks, a numerical measure is designed in this paper to assess the performance of given seeds. A sensitivity analysis is also conducted to test the impairing effect of several nodal importance metrics. Further, guided by the proposed measure, a memetic algorithm with the niching strategy has been devised to search for seeds with robust influential ability, termed NMA-RIM. Experiments on synthetic and real-world networks validate the competitiveness of NMA-RIM over existing algorithms, and the improved efficiency achieved by the niching strategy is also demonstrated. Meanwhile, the difference between selection results guided by the proposed measure and previous ones is shown via empirical analysis. As demonstrated by the obtained results, NMA-RIM outperforms existing approaches, and a better computational efficiency can be achieved. Competitive candidates can be provided for decision makers to solve information diffusion dilemmas.