Link Prediction in Multiplex Networks based on Topological and Latent Features

Currently, most of link prediction models are designed for single-layer networks, while for complex net-work systems in the real world, there are many different relationships between network entities. Thus, it is more advis-able to use multi-layer networks to model real-world complex systems. In this paper, we design a multi-layer network link prediction model to predict the missing edges in the target layer. The model not only extracts features of the target layer but also mines features from other auxiliary layers to help improve the prediction performance. In addition, rich network information can be obtained by combining both topological features and also latent features. In order to get latent features, the model decomposes the adjacency matrix of the network by using matrix factorization based on stochastic gradient descent, and each node can obtain a vector representation accordingly. Finally, topological features and latent representation are used to train a binary classifier to predict the missing edges in the target layer. The results on five real-world multiplex networks show that our model outperforms the compared models in this paper.