FASTEN:Uncovering Multiple Diffusion Networks Using the First-Hand Sharing Pattern

In our daily life rumors are spread among many people but diffusion processes and spreading paths behind rumors are usually hidden. The problem of finding this hidden process is getting more and more attention since after understanding the process, we can manipulate the diffusion speed of the process. For example, epidemiologists and the government can block a spread of a disease. Companies and goods sellers can facilitate adoptions of a product. To model the hidden information diffusion, it is usually assumed that information spreads in an underlying diffusion network where nodes, e.g. users, are receivers which adopt an information piece and edges, e.g. friendships, are transmission paths. In this work, we observe the pattern of information propagation that most of nodes are inclined to share the first-hand information. In other words, the virality of an information piece will generally decay as it becomes rephrased. We propose a generative model with the first-hand sharing patten (FASTEN), in order to infer the hidden networks and transmission rates between nodes. We further propose an efficient optimization method to infer the parameters of FASTEN. Experimental results show that FASTEN outperforms several state-of-the-arts algorithms on both synthetic and real datasets for network inference.