Influence Maximization in Big Networks: An Incremental Algorithm for Streaming Subgraph Influence Spread Estimation

Abstract

Influence maximization plays a key role in social network viral marketing. Although the problem has been widely studied, it is still challenging to estimate influence spread in big networks with hundreds of millions of nodes. Existing heuristic algorithms and greedy algorithms incur heavy computation cost in big networks and are incapable of processing dynamic network structures. In this paper, we propose an incremental algorithm for influence spread estimation in big networks. The incremental algorithm breaks down big networks into small subgraphs ad continuously estimate influence spread on these subgraphs as data streams. The challenge of the incremental algorithm is that subgraphs derived from a big network are not independent and MC simulations on each subgraph (defined as snapshots) may conflict with each other. In this paper, we assume that different combinations of MC simulations on subgraphs on subgraphs generate independent samples. In so doing, the incremental algorithm on streaming subgraphs can estimate influence spread with fewer simulations. Experimental results demonstrates the performance of the proposed algorithm.