LAUSR

Contagions such as information, rumors, infectious diseases, actions, and influence diffuse as cascades in large networks. Each contagion appears in some node and spreads through the nodes over the underlying network. In most cases, the network structure is hidden and we can only observe the times at which nodes are infected by contagions. So, inferring network structures and analyzing information diffusion processes are required in various domains. The vast majority of existing methods are parametric with the assumption that the diffusion patterns of contagions follow a particular distribution. In this paper, we propose a nonparametric method to infer the network topology, regardless of the contagion propagation model. We consider a static network and a set of given observed cascades, then, we try to infer not only the edges of the network; but also their strengths. First, we model the diffusion network as a Markov decision process (MDP), where the transition probabilities are assumed as pairwise transmission rates between the nodes. Later, a reinforcement learning paradigm is employed to solve this MDP. As compared to other methods, we do not make any assumption about the information diffusion pattern in our method; making it more general purpose. Experimental results on both synthetic and real datasets show that our method outperforms several state-of-the-art methods for different types of network structures and propagation models.