Benchmarking centrality heuristics for misinformation containment: a structure-dependent cross-network study

Abstract Misinformation spreads rapidly through online social networks, causing measurable harm to public health and undermining democratic discourse. The Influence Minimization problem, selecting nodes to immunise in order to contain such spread, is the structural inverse of the widely studied Influence Maximization problem, yet no systematic study has compared whether centrality heuristics transfer equally across both objectives. We benchmark eight shield-selection algorithms under an Independent Cascade model with node immunisation on Facebook Ego Network subgraphs from $$n=200$$ to $$n=2{,}000$$ nodes, across four intervention timings, a range of propagation probabilities and shield budgets, and two source models, and we replicate the core benchmark on three further networks (Twitter, Twitch, and an arXiv collaboration graph). On the modular Facebook network the algorithm ranking inverts relative to forward influence maximisation: Betweenness Shield, a mid-ranked forward seed-selection heuristic, becomes the single best containment method at every graph size, reaching a containment rate of $$\text {CR}=0.311$$ at $$n=2{,}000$$ and significantly outperforming all competitors, including a Monte-Carlo greedy baseline, with the advantage growing with graph size (all $$p<0.001$$ ). This advantage is, however, structure-dependent: it is decisive on the modular Facebook graph, reduces to a statistical tie on the dense Twitter and Twitch networks, and reverses in favour of degree-based blocking on the sparse collaboration network. Betweenness Shield also remains the best method across all tested intervention delays on Facebook. Bridge-blocking via betweenness centrality is thus the preferred containment strategy specifically on modular social networks with clear inter-community bridges.