Fig. 6: Average AUC scores for predicting links in the final layer (the target) in each of the 19 real-world networks.
From: Sequential stacking link prediction algorithms for temporal networks
a When the target layer is partially observed, we benchmark Top-Sequential-Stacking against Tensorial-SBM, Time-Series, and E-LSTM-D, and then also include these predictors with the stacking to compare against Ensemble-Sequential-Stacking. b When the target layer is completely unobserved, the Tensorial-SBM method cannot be used, so we benchmark against Time-Series and E-LSTM-D, and then include them as features in the stacking framework to benchmark with Ensemble-Sequential-Stacking. A random forest is used for the supervised stacking of methods.
