Social content based latent influence propagation model

With the proliferation of diversified social network services, understanding how the influence is propagated could help us apprehend the network evolution mechanism and the social impact of different kinds of information better. Most previous works have focused on the analysis of the influence propagation on the static network structure and the discovery of the subset of the most influential users. They fail to identify the user susceptibility delivered by user generated content. In this paper, we propose the InfoIBP (Influence propagation on Indian Buffet Process) model, a general framework for the latent influence propagation on social content with dynamic network structure, which based on the Indian buffet process. The influential users could be taken as the latent features in the social network and be found by different sampling algorithms based on numerical approximation. For the dynamic evolutional property of the network, hidden Markov model was adopted to describe the influence propagation in different time steps. A series of experiments for link prediction, preference prediction and running time evaluation are conducted on the DBLP and Digg datasets. The results show that the InfoIBP is more accurate and more efficient for modeling the latent influence propagation and discovering the influential users. It also can describe the dynamic evolutional property more comprehensively and achieve relatively accurate predictions for the future observations.