Abstract:Lots of neural network architectures have been proposed to deal with learning tasks on graph-structured data. However, most of these models concentrate on only node features during the learning process. The edge features, which usually play a similarly important role as the nodes, are often ignored or simplified by these models. In this paper, we present edge-featured graph attention networks, namely EGATs, to extend the use of graph neural networks to those tasks learning on graphs with both node and edge features. These models can be regarded as extensions of graph attention networks (GATs). By reforming the model structure and the learning process, the new models can accept node and edge features as inputs, incorporate the edge information into feature representations, and iterate both node and edge features in a parallel but mutual way. The results demonstrate that our work is highly competitive against other node classification approaches, and can be well applied in edge-featured graph learning tasks.
What problem does this paper attempt to address?
### Problems Addressed by the Paper
This paper aims to address a key issue in Graph Neural Networks (GNNs) when handling graph-structured data: most existing GNN models primarily focus on node features while neglecting edge features. Although edge features are equally important in many real-world tasks, and sometimes even more important than node features, existing models often ignore or simplify edge features.
Specifically, the paper proposes Edge-Featured Graph Attention Networks (EGATs) to extend the application scope of GNNs, enabling them to handle graph learning tasks that include both node and edge features. By improving the model structure and learning process, EGATs can accept node and edge features as input, integrate edge information into feature representations, and iteratively update node and edge features in a parallel but interrelated manner.
### Main Contributions
1. **Integration of Edge Features**: EGATs extend traditional Graph Attention Networks (GATs) to effectively utilize edge features.
2. **Symmetric Design**: The model is designed with a symmetric structure, allowing node and edge features to be updated in parallel and equivalently.
3. **Multi-Scale Fusion**: A multi-scale fusion strategy is adopted to concatenate features generated from different iterations, resulting in richer feature representations.
4. **Adaptation to Different Graph Characteristics**: EGATs can spontaneously handle the importance differences between node and edge features based on the preferences of the graph.
### Application Scenarios
The paper validates the effectiveness of EGATs in node classification tasks through experiments, particularly excelling in graphs with rich edge features. For example, in transaction networks, node labels may be highly correlated with transaction amounts, where edge features could have a greater impact on classification accuracy than node features.
### Experimental Results
- **Node-Sensitive Tasks**: On standard benchmark datasets such as Cora, Citeseer, and Pubmed, EGATs perform comparably to the current state-of-the-art models, even slightly outperforming them in some cases.
- **Edge-Sensitive Tasks**: On two transaction network datasets, Trade-B and Trade-M, EGATs show significant advantages, achieving classification accuracies of 92.0% and 85.4%, respectively.
### Conclusion
By integrating edge features, EGATs provide a more comprehensive approach to handling graph-structured data, especially in scenarios where edge features significantly impact task outcomes. This approach not only improves model performance but also offers new directions for research in Graph Neural Networks.