Scalable Link Prediction on Multidimensional Networks

Complex networks have been receiving increasing attention by the scientific community, also due to the availability of massive network data from diverse domains. One problem largely studied so far is Link Prediction, i.e. the problem of predicting new upcoming connections in the network. However, one aspect of complex networks has been disregarded so far: real networks are often multidimensional, i.e. multiple connections may reside between any two nodes. In this context, we define the problem of Multidimensional Link Prediction, and we introduce several predictors based on structural analysis of the networks. We present the results obtained on real networks, showing the performances of both the introduced multidimensional versions of the Common Neighbors and Adamic-Adar, and the derived predictors aimed at capturing the multidimensional and temporal information extracted from the data. Our findings show that the evolution of multidimensional networks can be predicted, and that supervised models may improve the accuracy of underlying unsupervised predictors, if used in conjunction with them.