High-Order Topology-Enhanced Graph Convolutional Networks for Dynamic Graphs

Understanding the evolutionary mechanisms of dynamic graphs is crucial since a basic characteristic real-world networks. The challenges modeling are as follows: (1) Real-world dynamics frequently characterized by group effects, which essentially emerge from high-order interactions involving groups entities. Therefore, pairwise revealed edges insufficient to describe complex systems. (2) graph data obtained real systems often noisy, and spurious can interfere with stability efficiency models. To address these issues, we propose topology-enhanced convolutional network for graphs. rationale behind it that symmetric substructure in graph, called maximal clique, reflect impacts on one hand, while not being readily disturbed links other hand. Then, utilize two independent branches model distinct influence effects. Learnable parameters used tune relative importance effects during process. We conduct link predictions datasets, including social citation Results show average improvements enhanced methods 68%, 15%, 280% over corresponding backbones across datasets. ablation study perturbation analysis validate effectiveness robustness proposed method. Our research reveals structures provide new perspectives studying highlight necessity employing higher-order topologies future.