A Survey on Edge-Aware Graph Learning Methods

doi:10.1109/TNSE.2025.3649386

A Survey on Edge-Aware Graph Learning Methods

2026-01-01

0

PRE

AI

José Augusto M. Nacif

R

Ricardo Ferreira

DOI：10.1109/TNSE.2025.3649386

Original

Original request for help

Save

Abstract

En 中文

Graph Neural Networks (GNNs) have gained popularity as an efficient choice for learning on graph-structured data. However, most methods are node or graph-centered, often overlooking valuable information that can be encoded in edge features and relations. In this survey, we present a comprehensive review and a novel taxonomy of Edge-Aware Graph Learning Methods, i.e., models that explicitly leverage edge information in the learning process. We trace the evolution of these methods from classical approaches through random walks to modern GNN architectures, including the emerging paradigm of Edge-Aware Graph Transformers. Through a comparative analysis, we demonstrate the consistent performance gains of these models over traditional node-centric approaches across a wide range of real-world applications and benchmarks. However, many challenges arise in this field. As such, we provide an explicit discussion of key limitations, particularly the scalability issues and computational overhead associated with many current architectures. Finally, by synthesizing the state-of-the-art and identifying open problems, this survey provides a clear roadmap to guide future research toward developing more efficient, scalable, and robust edge-aware models.

Keywords:

Machine learning

graphs

graph neural networks

edge-aware graph neural network

A Survey on Edge-Aware Graph Learning Methods

Abstract

Journal

Organization