Influence maximizing and local influenced community detection based on multiple spread model

Abstract

In independent cascade model, an active node has only one chance to activate its neighbors, while in reality an active node has many chances to activate its neighbors. We propose an influence diffusion model called multiple spread model, in which an active node has many activation chances. We prove that influence maximizing problem with the proposed model is submodular and monotone, which means greedy algorithm provides (1-1/e) approximation to optimal solution. However, computation time costs much due to Monte Carlo simulation in greedy algorithm. We propose a two-phase method which leverages community information to find seeds. In order to evaluate influence of a particular node, we also propose a definition of local influenced community as well as an algorithm called LICD to detect local influenced community. Experiments show that the proposed model and algorithms are both efficient and effective in problems of influence maximizing and local influenced community detection. © 2011 Springer-Verlag.