Scalable and Parallel Processing of Influence Maximization for Large-Scale Social Networks

Abstract

Influence maximization is a problem of finding a small subset of nodes as seeds in a social network such that the total influence of this subset of nodes for disseminating a message in the social network can be maximized. The problem has been extensively investigated in recent years and many influence maximization algorithms have been proposed. However, all of the existing algorithms are sequentially executed algorithms. It would take long time if they run on large-scale social networks. In this paper, we study parallel algorithms for two influence maximization problems in large-scale social networks: influence maximization without budget limitation and influence maximization with limited budget. We propose two parallel algorithms, Community-based Max Degree (CMD) algorithm and Max Degree Cost Ratio (MDCR) algorithm, respectively for the two problems. Both algorithms can run in parallel on Hadoop platform. Experiments are conducted for various sizes of social networks. The results show that our algorithms are scalable and outperform the common heuristic algorithms.