Current content moderation efforts targeting individual toxic users often fail to disrupt dense, toxic communities. We formalize toxicity containment as a network optimization problem where focal toxic structures (FTSs)-high-degree subgraphs with mean toxicity \(\ge 0.68\) -serve as toxicity propagation super-spreaders in modular networks. Our WFSA-IP framework combines weighted focal structure analysis with integer programming to select FTSs for intervention, formulated as a constrained Multiple-Knapsack Problem maximizing disruption efficiency \(\phi _i/|F_i|\) under operational constraints: budget \(\le \) 5100 users, impact \(\ge \) 12%, overlap \(\le \) 0.3. Validation on Telegram (324,769 users, Russia-Ukraine discourse) shows IP-selected FTSs achieve 77.5% toxicity removal, 6.6% network fragmentation, and 3.3% connection disruption vs. 32.0%, 3.5%, and 1.4% (IP-unselected FTSs) and 15.1%, 1.0%, and 0.6% (Influential Toxic Individuals), yielding \(4.2\times \) efficiency improvement. Cross-platform validation on Twitter/X StopTheSteal Data (195,658 users) confirms \(3.8\times \) efficiency gains. Statistical analysis reveals very large effect sizes (Cohen’s \(d = 15.3\) , \(p < 0.001\) ). The WFSA-IP framework transforms content moderation from a reactive, individual-targeting approach to a proactive, structural intervention, while preserving network integrity and advancing network science by demonstrating how targeted subgraph removal alters the dynamics of contagion in complex systems.