IMPC: Influence maximization based on multi-neighbor potential in community networks

The study of influence maximization (IM) has attracted many scholars in recent years due to its import practical values. Given a social network, it aims at finding a subset of k individuals as seed nodes which can trigger the maximum influence cascade through the network under a predefined diffusion model. Kempe et al. first formulated influence maximization as a discrete optimization problem and proved its NP-hardness and submodularity, based on which they further proposed a greedy approach with guaranteed solution accuracy. Unfortunately, the greedy algorithm was also known for its extremely low time efficiency. To solve this problem more efficiently, many research works were proposed in recent years. However, these studies either make sacrifices in solution accuracy or require huge memory consumption. Besides, only a handful of research works can handle mega-scale networks with millions of nodes and edges. To solve this problem both efficiently and effectively, in this paper we propose IMPC: an influence maximization framework based on multi-neighbor potential in community networks. In our approach the influence diffusion process is divided into two phases: (i) multi-neighbor potential based seeds expansion; and (ii) intra-community influence propagation. Based on this framework we derive an objective function to evaluate the overall influence as a combination of the influence during the two phases. We theoretically prove that the objective function is submodular and design an efficient greedy algorithm to find the seed nodes. We evaluate the performance of our framework on eight real-world networks which scale up to millions of nodes and hundreds of millions of edges. Experimental results show that our approach can significantly outperform other state-of-the-art algorithms in terms of time and space efficiency with no compromise on solution accuracy. (C) 2018 Elsevier B.V. All rights reserved.