Adaptive Rumor Suppression Mechanism Based on Directed Graph Convolutional Networks

Online social networks enable unverified information to reach large audiences quickly, and rumor circulation can undermine informed decision-making and trust. Multiple rumor topics may spread at the same time, differ in their diffusion patterns, and reach overlapping users. These conditions complicate the allocation of limited resources between blocking and clarification. This study formulates an adaptive multi-type rumor suppression problem and develops a multi-type rumor linear threshold (MTRLT) model with cross-topic promoting influence and separate contact and decision stages. To select intervention candidates, a directed graph convolutional network combines network structure, user behavior, and topic inclination; its topic-specific node scores are trained with binary cross-entropy. A hybrid strategy allocates blocking and clarification actions under a common budget. Simulation experiments on networks derived from Digg and Weibo evaluate terminal rumor states under paired source conditions. The formal validation shows that DGCN ranking outperforms deterministic Top-Degree and PageRank rankings under a common Hybrid backend, but not a randomized Degree ranking, and that the benefits of learning, hybrid allocation, feedback rounds, direction mixing, and feature groups depend on the network setting. The study provides a directed-network framework for modeling topic-dependent rumor diffusion and evaluating budgeted interventions without claiming universal superiority across strategies or networks.