Cascade with varying activation probability model for influence maximization in social networks

Activation probability is a key parameter in information diffusion models and has been observed to be varying with history activations in many empirical studies. However, such phenomenon has not been incorporated in the diffusion models applied in Influence Maximization Problem. In this paper, we first conduct empirical analyses on the large scale dataset collected from a popular online social network to demonstrate the variation. Then we propose the Cascade with Varying Activation Probability (CVAP) model and validate its accuracy by extensive simulation experiments. Moreover, we prove the submodularity of CVAP model, which guarantees that greedy algorithm can achieve 1 - 1/e optimality when solving the influence maximization problem.