Finding robust and influential nodes on directed networks using a memetic algorithm

Abstract

The influence maximization problem is a research hotspot on social networks, which involves selecting a set of nodes as seeds to maximize the influence spread. The robust influence maximization problem is a further extension of the influence maximization problem considering the external factors pertaining to the propagation process. The current research mainly includes how to select the most influential nodes through algorithms, or how to improve the robustness of the network structure. However, these researches lack the application of the propagation models in directed networks, and also ignore the networked robustness on the information transmission. Therefore, this paper focuses on analyzing the characteristics of directed networks in the information propagation process, and further solve the robust influence maximization problem considering the directionality. First, a robust influence evaluation factor RSD is designed to assess the robust influence ability of each node on directed networks. Under the guidance of this evaluation factor, a problem-directed Memetic algorithm is developed to solve the robust influence maximization problem, named MA-RIMD. The characteristics of directed networks are considered involving with the genetic operators. The empirical analysis on several synthetic and practical networks indicates that MA-RIMD can select robust and influential seeds from directed networks, outperforming baseline approaches. Based on experiments conducted on three synthetic networks and two real-world networks, our proposed algorithm exhibits a performance improvement up to 15% over baseline approaches.