[Paper Review] Graph Neural Networks: Methods, Applications, and Opportunities

Lilapati Waikhom, Ripon Patgiri|arXiv (Cornell University)|Aug 24, 2021

Advanced Graph Neural Networks192 references31 citations

TL;DR

A comprehensive survey of graph neural networks (GNNs) across learning settings, providing taxonomy, architectures, datasets, applications, challenges, and future directions.

ABSTRACT

In the last decade or so, we have witnessed deep learning reinvigorating the machine learning field. It has solved many problems in the domains of computer vision, speech recognition, natural language processing, and various other tasks with state-of-the-art performance. The data is generally represented in the Euclidean space in these domains. Various other domains conform to non-Euclidean space, for which graph is an ideal representation. Graphs are suitable for representing the dependencies and interrelationships between various entities. Traditionally, handcrafted features for graphs are incapable of providing the necessary inference for various tasks from this complex data representation. Recently, there is an emergence of employing various advances in deep learning to graph data-based tasks. This article provides a comprehensive survey of graph neural networks (GNNs) in each learning setting: supervised, unsupervised, semi-supervised, and self-supervised learning. Taxonomy of each graph based learning setting is provided with logical divisions of methods falling in the given learning setting. The approaches for each learning task are analyzed from both theoretical as well as empirical standpoints. Further, we provide general architecture guidelines for building GNNs. Various applications and benchmark datasets are also provided, along with open challenges still plaguing the general applicability of GNNs.

Motivation & Objective

• Define basic terminologies and variants of graphs and graph-based tasks.
• Present a taxonomy of GNNs for each learning setting (supervised, unsupervised, semi-supervised, self-supervised).
• Analyze GNN methods from theoretical and empirical perspectives across learning settings.
• Provide general guidelines for designing GNN architectures.
• Survey datasets, applications, and open challenges to guide future research.

Proposed method

• Describe graph representations and message-passing frameworks used in GNNs.
• Develop taxonomies for GNN variants across learning settings (supervised, unsupervised, semi-supervised, self-supervised).
• Classify tasks at node, edge, graph, and subgraph levels and discuss corresponding methods.
• Summarize datasets, applications, and benchmarking practices in GNN research.
• Outline architecture guidelines and design considerations for scalable and robust GNNs.
• Discuss open challenges and future research directions in depth.

Experimental results

Research questions

• RQ1 What are the prevailing GNN methods across different learning settings (supervised, unsupervised, semi-supervised, self-supervised)?
• RQ2 How do graph variants and task hierarchies influence GNN design and performance across domains?
• RQ3 What theoretical and empirical insights explain GNN successes and limitations, and what are future research directions?
• RQ4 Which datasets and benchmarks best reflect the capabilities and challenges of GNNs across tasks?
• RQ5 What guidelines can be established for constructing effective and scalable GNN architectures?

Key findings

• Provides a new taxonomy of GNNs tailored to each learning setting.
• Offers a theoretical and empirical analysis of GNN approaches across tasks and domains.
• Delivers general architecture guidelines for building GNNs.
• Documents widely-used datasets and applications in graph-based learning.
• Identifies open challenges and proposes future research directions in depth (depth, scalability, higher-order structures, robustness).