Hypergraph-Based Dynamic Graph Node Classification

TL;DR

The paper tackles node classification on discrete dynamic graphs, where topology, attributes, and labels evolve over time, and shows that traditional RNN or self-attention approaches struggle to capture diverse temporal dynamics. It introduces HYDG, a Hypergraph-Based Multi-granularity Dynamic Graph Node Classification framework that jointly models individual-level and group-level spatio-temporal dependencies: individual-level hypergraphs link each node to K temporal neighbors across time, while group-level hypergraphs cluster same-label nodes to capture multi-granularity class dynamics; a Hypergraph Neural Network propagates information across both hypergraph structures to produce fused representations. The model optimizes two losses, L_in and L_group, weighted by alpha and beta, and uses the trained network to predict node labels, with group-level hypergraphs providing data augmentation for unseen test sets. Experiments on five real dynamic graph datasets across two backbones (GCN and GraphSAGE) demonstrate that HYDG consistently outperforms strong baselines in both accuracy and Macro-AUC, validating the effectiveness and robustness of the multi-granularity hypergraph approach for dynamic graphs.

Abstract

Node classification on static graphs has achieved significant success, but achieving accurate node classification on dynamic graphs where node topology, attributes, and labels change over time has not been well addressed. Existing methods based on RNNs and self-attention only aggregate features of the same node across different time slices, which cannot adequately address and capture the diverse dynamic changes in dynamic graphs. Therefore, we propose a novel model named Hypergraph-Based Multi-granularity Dynamic Graph Node Classification (HYDG). After obtaining basic node representations for each slice through a GNN backbone, HYDG models the representations of each node in the dynamic graph through two modules. The individual-level hypergraph captures the spatio-temporal node representations between individual nodes, while the group-level hypergraph captures the multi-granularity group temporal representations among nodes of the same class. Each hyperedge captures different temporal dependencies of varying lengths by connecting multiple nodes within specific time ranges. More accurate representations are obtained through weighted information propagation and aggregation by the hypergraph neural network. Extensive experiments on five real dynamic graph datasets using two GNN backbones demonstrate the superiority of our proposed framework.

Figures (3)

• Figure 1: An illustrative example of interest prediction in dynamic social graphs involves predicting users' interest tags at a specific time based on historical data. In this network, both user connections and interest tags evolve over time, and our goal is to forecast users' interest tags using information from the interest network.
• Figure 2: The detailed architecture of HYDG consists of three components: graph embedding(blue and orange represent different labels), hypergraph construction (for better visualization, we only illustrate the binary clustering approach),and hypergraph propagation.
• Figure 3: Ablation study results of HYDG on node classification tasks across five datasets.