Probing the fitness landscape of the influential nodes for the influence maximization problem in social networks

Abstract

Influence Maximization (IM) is a key issue of information dissemination and has been proved to be an NP-hard problem. However, traditional methods always suffer from low efficiency, poor scalability, and tend to fall into local optima. Probing the promising distribution regions of the potential influential nodes from the macroscopic perspective is necessary and helpful in understanding the influence propagation. To address such challenges, this paper makes attempt to depict the fitness landscape distribution of the expected influence of the social individuals in the network from a novel perspective. An entropy measure is introduced as a decision criterion and a fitness landscape-guided differential evolution optimization (FLDE) is proposed. Firstly, the distribution of the potential solution regions is depicted by characterizing the fitness landscape designed specially for IM problem. Next, a guiding strategy based on the fitness landscape is conceived to drive the differential evolution towards more promising solution regions by avoiding the entrapment in local optima. Experiments conducted on six real social networks and three synthetic networks indicate that the FLDE outperforms the state-of-the-art baselines by an average of 16% in influence spread and shows strong scalability when dealing with different types of networks.