Inferring links in cascade through hawkes process based diffusion model

Abstract

Data of information cascade in social network is always incomplete with missing information of the links between nodes. This paper proposes a generative probabilistic model to infer links using the observation data. Comparing to existing methods, we take consideration of differences of links. And we are also in view of recurrent events and influence from outside of the cascade. Our hawkes process based diffusion model (HPBDM) is testified to precede the prior models in the aspect of inferring links on synthetic data and real data. We also modify the HPBDM by adding time threshold and build up modified hawkes process based diffusion model (MHPBDM). Conducting experiment on real data with MHPBDM, we discover that it is more suitable for some kinds of information whose time interval for information cascade is long.