Path-wise Cascading Probabilistic Description for Information Diffusion in Networks
Qi Zhang, Rui Li and Rui Mao
We consider the information diffusion issue in networks, where a pathwise continuous transmission model and a partially-observed cascade model are proposed in this paper to probabilistically describe the diffusion processes. In networks, a diffusion process begins from a source vertex, and then propagates through the whole network. Each edge transmits the information by a certain likelihood, which is based on the edge parameter. Since in many real applications, merely partial observations of the diffusion process can be obtained, describing the temporal dynamic of the partially-observed continuous diffusion processes becomes challenging and it suffers the difficulty of computing the transmission likelihood through uncertain transmission paths. To solve this problem, we propose a path-wise continuous transmission model to unveil the probabilistic dynamics through potential transmission paths. Based on it, a partially-observed cascading probabilistic model is presented by using the introduced path-wise transmission model to capture the likelihood of the diffusion process under partial observations. Compared with the traditional cascade model, the proposed models can be better applied in partial-observation tasks. Simulations under both synthetic and real networks are conducted to evaluate the proposed probabilistic models.
Keywords: Social networks, information diffusion, partial observation, continuous transmission