Abstract:In this paper, we study scalable algorithms for influence maximization with general marketing strategies (IM-GMS), in which a marketing strategy mix is modeled as a vector $mathbf{x}=(x_1, ldots, x_d)$ and could activate a node $v$ in the social network with probability $h_v(mathbf{x})$. The IM-GMS problem is to find the best strategy mix $mathbf{x}^*$ that maximize the influence spread due to influence propagation from the activated seeds, subject to the budget constraint that $sum_{jin [d]} x_j le k$. We adapt the scalable reverse influence sampling (RIS) approach and design a scalable algorithm that provides a $(1-1/e -varepsilon)$ approximate solution (for any $varepsilon u003e 0$), with running time near-linear in the network size. We further extend IM-GMS to allow partitioned budget constraint, and show that our scalable algorithm provides a $(1/2-varepsilon)$ solution in this case. Through extensive experiments, we demonstrate that our algorithm is several orders faster than the Monte Carlo simulation based hill-climbing algorithm, and also outperforms other baseline algorithms proposed in the literature.

Scalable Influence Maximization with General Marketing Strategies.