Journal Article
Knowledge Graphs
Featured

A Comprehensive Survey on Graph Neural Networks

Zonghan Wu(University of Technology Sydney), Shirui Pan(Monash University), Fengwen Chen(University of Technology Sydney), Guodong Long(University of Technology Sydney), Chengqi Zhang(University of Technology Sydney), Philip S. Yu(University of Illinois Chicago)
March 24, 2020IEEE Transactions on Neural Networks and Learning Systems11,563 citations
arXivPDFDOI

12k

Citations

514

Influential Citations

IEEE Transactions on Neural Networks and Learning Systems

Venue

2020

Year

Abstract

Deep learning has revolutionized many machine learning tasks in recent years, ranging from image classification and video processing to speech recognition and natural language understanding. The data in these tasks are typically represented in the Euclidean space. However, there is an increasing number of applications, where data are generated from non-Euclidean domains and are represented as graphs with complex relationships and interdependency between objects. The complexity of graph data has imposed significant challenges on the existing machine learning algorithms. Recently, many studies on extending deep learning approaches for graph data have emerged. In this article, we provide a comprehensive overview of graph neural networks (GNNs) in data mining and machine learning fields. We propose a new taxonomy to divide the state-of-the-art GNNs into four categories, namely, recurrent GNNs, convolutional GNNs, graph autoencoders, and spatial-temporal GNNs. We further discuss the applications of GNNs across various domains and summarize the open-source codes, benchmark data sets, and model evaluation of GNNs. Finally, we propose potential research directions in this rapidly growing field.

Analysis

Why This Paper Matters

Graph Neural Networks (GNNs) have emerged as a critical tool for learning on non-Euclidean data, such as social networks, molecular structures, and knowledge graphs. This survey, published in IEEE TNNLS in 2020, provides a structured taxonomy that has helped researchers and practitioners navigate the rapidly growing field. With over 11,500 citations, it is one of the most influential papers in graph deep learning, serving as a standard reference for both newcomers and experts.

The paper's timing was crucial: it captured the state of the art just as GNNs were transitioning from niche research to mainstream adoption in areas like drug discovery, recommendation systems, and traffic forecasting. By organizing methods into four clear categories, it enabled systematic comparison and inspired subsequent innovations.

Technical Contributions

The paper's main innovation is its taxonomy, which divides GNNs into:

  • Recurrent GNNs: Early models that process graph nodes iteratively using recurrent architectures (e.g., GraphESN, GGNN).
  • Convolutional GNNs: Models that generalize convolution to graphs, including spectral methods (e.g., ChebNet, GCN) and spatial methods (e.g., GraphSAGE, GAT).
  • Graph Autoencoders: Unsupervised learning frameworks for node embedding and graph generation (e.g., VGAE, ARGA).
  • Spatial-Temporal GNNs: Models that handle dynamic graphs with temporal dependencies (e.g., STGCN, DCRNN).

Additionally, the survey compiles open-source code repositories, benchmark datasets (e.g., Cora, Citeseer, Pubmed, PPI), and evaluation protocols, providing a practical resource for implementation.

Results

As a survey, the paper does not present new experimental results. However, it systematically reviews performance of key models on standard benchmarks. For example, it notes that GCN achieves 81.5% accuracy on Cora, GraphSAGE achieves 93.0% F1 on PPI, and GAT achieves 83.0% accuracy on Cora. The survey also discusses computational complexity and scalability trade-offs between spectral and spatial methods.

Significance

This survey has had a profound impact on the AI field by democratizing access to GNN knowledge. It has been used as a textbook-style introduction in university courses and industry tutorials. The taxonomy it proposed has been widely adopted in subsequent surveys and research papers. By highlighting open challenges such as scalability, interpretability, and dynamic graphs, it has shaped the research agenda for years to come. For practitioners at Neura Market, this paper remains essential reading for understanding the foundations of graph-based deep learning.

graph-neural-networksdeep-learningsurveyknowledge-graphsconvolutional-gnngraph-autoencodersspatial-temporal-gnn
BenchmarksNeural Networks

Related Papers