Big social network influence maximization via recursively estimating influence spread

Influence maximization aims to find a set of highly influential nodes in a social network to maximize the spread of influence. Although the problem has been widely studied, it is still challenging to design algorithms to meet three requirements simultaneously, i.e., fast computation, guaranteed accuracy, and low memory consumption that scales well to a big network. Existing heuristic algorithms are scalable but suffer from unguaranteed accuracy. Greedy algorithms such as CELF [1] are accurate with theoretical guarantee but incur heavy simulation cost in calculating the influence spread. Moreover, static greedy algorithms are accurate and sufficiently fast, but they suffer extensive memory cost. In this paper, we present a new algorithm to enable greedy algorithms to perform well in big social network influence maximization. Our algorithm recursively estimates the influence spread using reachable probabilities from node to node. We provide three strategies that integrate memory cost and computing efficiency. Experiments demonstrate the high accuracy of our influence estimation. The proposed algorithm is more than 500 times faster than the CELF algorithm on four real world data sets.