Fig. 2: Comparison of link prediction performance in the partially-observed setting using temporal common neighbors (vertical) versus sequential common neighbors (horizontal).
From: Sequential stacking link prediction algorithms for temporal networks
Performance is quantified here by the area under the receiver operating characteristic curve (AUC), from each run over 5-fold cross-validation on the target layer, on each of a 19 real-world temporal networks, b 45 random realizations from degree-corrected community-label temporal stochastic block models (T-SBM) at different parameters, and (c) 45 random edge-correlated T-SBM realizations at different parameters. (See Methods and Fig. 4 for the 45 different parameter combinations in each.) The diagonal line (y = x) indicates equal performance.
