Fig. 4: Average AUC for partially-observed target layers on synthetic datasets.
From: Sequential stacking link prediction algorithms for temporal networks
(a) community-label T-SBM, and (b) edge-correlated T-SBM networks, each with 200 nodes and 10 layers. Each row shows results for a different fraction of random edges, μ (see Methods), at different copy probabilities, p (of community labels in a, of edges in b), and numbers of communities, k. Dashed lines represent the theoretical maximum average link prediction performance achievable by an oracle that knows the full T-SBM specification. Colors indicate different link prediction methods (see legend). A random forest is used for supervised stacking of methods.
