Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graph Neural Networks: A Review of Methods and Applications

Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, Maosong Sun

TL;DR

Graph Neural Networks (GNNs) are reviewed through a designer-friendly pipeline that begins with identifying graph structure, graph type/scale, and task-specific loss, then assembling propagation, sampling, and pooling modules. The survey details spectral and spatial convolution variants, recurrent and skip-connection designs, and sampling/pooling strategies, and extends to directed, heterogeneous, dynamic, hypergraph, signed, and large graphs. It also covers training-settings variants (auto-encoders, contrastive learning), a design example (GPT-GNN), theoretical analyses (signal processing, generalization, expressivity, invariance, transferability, label efficiency), and a broad spectrum of applications in structural and non-structural domains. Four open problems are highlighted: robustness to adversarial attacks, interpretability, graph pretraining, and handling complex graph structures. Overall, the paper provides a thorough, modular viewpoint that aids researchers and practitioners in selecting and designing GNN architectures for diverse tasks.

Abstract

Lots of learning tasks require dealing with graph data which contains rich relation information among elements. Modeling physics systems, learning molecular fingerprints, predicting protein interface, and classifying diseases demand a model to learn from graph inputs. In other domains such as learning from non-structural data like texts and images, reasoning on extracted structures (like the dependency trees of sentences and the scene graphs of images) is an important research topic which also needs graph reasoning models. Graph neural networks (GNNs) are neural models that capture the dependence of graphs via message passing between the nodes of graphs. In recent years, variants of GNNs such as graph convolutional network (GCN), graph attention network (GAT), graph recurrent network (GRN) have demonstrated ground-breaking performances on many deep learning tasks. In this survey, we propose a general design pipeline for GNN models and discuss the variants of each component, systematically categorize the applications, and propose four open problems for future research.

Graph Neural Networks: A Review of Methods and Applications

TL;DR

Graph Neural Networks (GNNs) are reviewed through a designer-friendly pipeline that begins with identifying graph structure, graph type/scale, and task-specific loss, then assembling propagation, sampling, and pooling modules. The survey details spectral and spatial convolution variants, recurrent and skip-connection designs, and sampling/pooling strategies, and extends to directed, heterogeneous, dynamic, hypergraph, signed, and large graphs. It also covers training-settings variants (auto-encoders, contrastive learning), a design example (GPT-GNN), theoretical analyses (signal processing, generalization, expressivity, invariance, transferability, label efficiency), and a broad spectrum of applications in structural and non-structural domains. Four open problems are highlighted: robustness to adversarial attacks, interpretability, graph pretraining, and handling complex graph structures. Overall, the paper provides a thorough, modular viewpoint that aids researchers and practitioners in selecting and designing GNN architectures for diverse tasks.

Abstract

Lots of learning tasks require dealing with graph data which contains rich relation information among elements. Modeling physics systems, learning molecular fingerprints, predicting protein interface, and classifying diseases demand a model to learn from graph inputs. In other domains such as learning from non-structural data like texts and images, reasoning on extracted structures (like the dependency trees of sentences and the scene graphs of images) is an important research topic which also needs graph reasoning models. Graph neural networks (GNNs) are neural models that capture the dependence of graphs via message passing between the nodes of graphs. In recent years, variants of GNNs such as graph convolutional network (GCN), graph attention network (GAT), graph recurrent network (GRN) have demonstrated ground-breaking performances on many deep learning tasks. In this survey, we propose a general design pipeline for GNN models and discuss the variants of each component, systematically categorize the applications, and propose four open problems for future research.

Paper Structure

This paper contains 67 sections, 26 equations, 7 figures, 8 tables.

Table of Contents

  1. R T I C L E I N F O
  2. Introduction
  3. General design pipeline of GNNs
  4. Find graph structure
  5. Specify graph type and scale
  6. Design loss function
  7. Build model using computational modules
  8. Instantiations of computational modules
  9. Propagation modules - convolution operator
  10. Spectral approaches
  11. Basic spatial approaches
  12. Attention-based spatial approaches
  13. General frameworks for spatial approaches
  14. Propagation modules - recurrent operator
  15. Convergence-based methods
  16. ...and 52 more sections

Figures (7)

  • Figure 1: Left: image in Euclidean space. Right: graph in non-Euclidean space.
  • Figure 2: The general design pipeline for a GNN model.
  • Figure 3: An overview of computational modules.
  • Figure 4: An overview of variants considering graph type and scale.
  • Figure 5: An overview of methods with unsupervised loss.
  • ...and 2 more figures