Online rumor detection is essential for maintaining trustworthy social platforms. With the rapid growth of misinformation on social networks, developing automated models that exploit both semantic and structural cues has become a key research direction. Traditional rumor detection approaches mainly rely on text content or user profiles, which can be easily manipulated and often fail to capture the underlying diffusion behaviors of information. In contrast, propagation structures convey richer and more stable cues for identifying deceptive information. However, modeling these complex retweeting graphs is challenging due to their irregular topology and dynamic edge relationships. To address this, we propose a novel Edge-Weighted Hypergraph Neural Network framework for structure-aware rumor detection. The model constructs dynamic retweeting graphs and assigns adaptive edge weights based on interaction intensity and user influence. A hypergraph fusion layer aggregates higher-order connections among multiple propagation subgraphs, while a joint attention fusion mechanism integrates textual semantics, user behavior, and topological patterns. Experiments on four benchmark datasets — Weibo, Twitter15, Twitter16, and PHEME- demonstrate the superiority of Edge-Weighted Hypergraph Neural Network, achieving accuracies of 93.9%, 82.8%, 82.9%, and 88.9%, respectively, outperforming state-of-the-art models such as RvNN, Bi-GCN, and DA-GCN. These findings confirm that the proposed framework provides a robust and generalizable solution for early and accurate rumor detection in online environments. • Rumor detection in online social networks helps to maintain the user’s trust in the information shared on their platforms. • Binary tree structure for organizing and efficiently traversing the retweeting graph to reduce the computational complexity and improving scalability. • Meta graph topology enriches the retweeting graph with additional contextual information. • Extract the graphlets (small graphs) to identify and learn local structure patterns indicative of rumor propagation. • Hyperedge neural networks to model complex interactions involving multiple nodes, capturing high-order dependencies. • Assigns weights to edges based on user influence, engagement, and temporal factors, accurately measuring interaction importance and enhancing the model’s effectiveness in identifying key players in rumor detection.
Kondamudi et al. (Fri,) studied this question.