EdgeGFL: Rethinking Edge Information in Graph Feature Preference Learning

TL;DR

EdgeGFL addresses the gap in graph neural networks by incorporating multidimensional edge features into the message-passing framework, enabling edge embeddings to actively refine node representations. It introduces a feature projection module, edge-type driven initialization, and a feature-preference based propagation mechanism with residual-augmented aggregation, culminating in an efficient, edge-aware GNN architecture. Across four real-world heterogeneous graphs, EdgeGFL achieves state-of-the-art performance in node classification and clustering, with ablation studies demonstrating the critical role of edge-type encoding and edge feature initialization, and time analyses confirming scalable training. This work advances graph representation learning by tightly coupling edge and node information, improving non-local structure capture and high-order feature modeling for complex graph data.

Abstract

Graph Neural Networks (GNNs) have significant advantages in handling non-Euclidean data and have been widely applied across various areas, thus receiving increasing attention in recent years. The framework of GNN models mainly includes the information propagation phase and the aggregation phase, treating nodes and edges as information entities and propagation channels, respectively. However, most existing GNN models face the challenge of disconnection between node and edge feature information, as these models typically treat the learning of edge and node features as independent tasks. To address this limitation, we aim to develop an edge-empowered graph feature preference learning framework that can capture edge embeddings to assist node embeddings. By leveraging the learned multidimensional edge feature matrix, we construct multi-channel filters to more effectively capture accurate node features, thereby obtaining the non-local structural characteristics and fine-grained high-order node features. Specifically, the inclusion of multidimensional edge information enhances the functionality and flexibility of the GNN model, enabling it to handle complex and diverse graph data more effectively. Additionally, integrating relational representation learning into the message passing framework allows graph nodes to receive more useful information, thereby facilitating node representation learning. Finally, experiments on four real-world heterogeneous graphs demonstrate the effectiveness of theproposed model.

Figures (8)

• Figure 1: Illustration of the difference between information preferences and attention mechanisms in a message passing framework. (a) The edge feature representation in ordinary GCN; (b) The edge feature representation in our proposed EdgeGFL.
• Figure 2: Illustration of the $l$-th layer of the EdgeGFL model. The EdgeGFL model achieves the final representation of the target node in a heterogeneous graph through node and edge embeddings, information propagation, and aggregation processes.
• Figure 3: Visualization of embedding on ACM. Nodes with different labels are differentiated by colors.
• Figure 4: Visualization of embedding on DBLP. Nodes with different labels are differentiated by colors.
• Figure 5: Experimental results of ablation study

Theorems & Definitions (1)

• Definition 1: Feature Preference Aware Message-passing, FPAMP