Latent Feature Independent Cascade Model for Social Propagation

People share various types of information including opinions on hot topics, bookmarking activity and rumors via online communities. To make it possible to predict future trends in online communities, it is important that we develop a model of information diffusion through social networks and a method for estimating its parameters. In this paper, we present a latent feature independent cascade model, which can effectively estimate diffusion probabilities by capturing the influences between latent communities. In particular, we incorporate two types of latent features for each node. The first represents the features as a sender and the second represents the features as a receiver. We demonstrate experimentally that the proposed model can estimate the diffusion probabilities more accurately than commonly used methods. We also show the effectiveness of the proposed model for estimating information spread.