Inferring Diffusion Network on Incomplete Cascade Data

Inferring the diffusion network based on observed cascades is fundamental and of interest in the field of information diffusion on the network. All the existing methods which aim to infer the network assume that the cascades are complete without any missing nodes. In real world, not every infection between nodes can be easily observed or acquired. As a result, there are some missing nodes in the real cascades, which indicates that the observed cascades are incomplete and makes more challenges for solving this problem. Being able to recover the incomplete cascades is critical to us since inferred network based on the incomplete cascades can be inaccurate. In this paper, we tackle the problem by developing a two-stage framework, which finds the paths that contain the missing nodes at first and then estimate the location and infection time of missing nodes. Experiments on real and synthetic data show the accuracy of our algorithm to finding the missing node on the network, as well as the infection time of the missing nodes.