Dynamic Link Prediction Using Restricted Boltzmann Machine

We consider the problem of link prediction in dynamic networks under the condition of a set of snapshots of the networks. To address the nonlinear transitional patterns in network structures, we propose an approach that incorporates the historical linkage and neighboring information into the restricted Boltzmann machine (RBM) model by adding temporal and neighboring connections between the hidden and visible layers. Based on this, we give an algorithm to train the model parameters. Experiments on synthetic dynamic networks verify the effectiveness of our method.