Fig. 13

Increasing link-prediction error in a messenger-app network (\(\text {OAS}_{\text {tg}}\) for 1281 nodes). \(\text {OAS}_{\text {tg}}\) as a function of false negative rate in the UCI network and all nodes have infection thresholds 0.2, 0.4, 0.6, 0.8. Budgets are sufficient for traditional greedy to give a complete cascade in \(G\). Budgets are (5, 34, 131, 371). Since traditional greedy is very slow, resolution and replication are reduced: computing cascade margins for this figure required over 27 million simulations of spread through a 1281 node graph. Five replications are conducted at each of eleven \(p_{\text {neg}}\) values