[논문 리뷰] Graph Data Augmentation for Graph Machine Learning: A Survey

Tong Zhao, Jin, Wei외 6명|arXiv (Cornell University)|2022. 02. 17.

Advanced Graph Neural Networks인용 수 42

한 줄 요약

포괄적인 그래프 데이터 증강(GDA) 방법, 분류 체계, 학습 기반과 규칙 기반 접근 방식, 그리고 그래프 ML의 향후 과제에 대한 종합적 고찰.

ABSTRACT

Data augmentation has recently seen increased interest in graph machine learning given its demonstrated ability to improve model performance and generalization by added training data. Despite this recent surge, the area is still relatively under-explored, due to the challenges brought by complex, non-Euclidean structure of graph data, which limits the direct analogizing of traditional augmentation operations on other types of image, video or text data. Our work aims to give a necessary and timely overview of existing graph data augmentation methods; notably, we present a comprehensive and systematic survey of graph data augmentation approaches, summarizing the literature in a structured manner. We first introduce three different taxonomies for categorizing graph data augmentation methods from the data, task, and learning perspectives, respectively. Next, we introduce recent advances in graph data augmentation, differentiated by their methodologies and applications. We conclude by outlining currently unsolved challenges and directions for future research. Overall, our work aims to clarify the landscape of existing literature in graph data augmentation and motivates additional work in this area, providing a helpful resource for researchers and practitioners in the broader graph machine learning domain. Additionally, we provide a continuously updated reading list at https://github.com/zhao-tong/graph-data-augmentation-papers.

연구 동기 및 목표

• Clarify the motivation for graph data augmentation (GDA) in graph neural networks (GNNs).
• Provide a structured taxonomy of GDA methods from data, task, and learning perspectives.
• Summarize representative GDA techniques and their applications across node-, edge-, and graph-level tasks.
• Identify unsolved challenges and propose directions for future research in GDA.

제안 방법

• Present three orthogonal taxonomies: operated data modality (structure, feature, label augmentations), downstream task (node/edge/graph), and learning paradigm (rule-based vs. learned).
• Describe rule-based GDA techniques categorized as data removal, data addition, and data manipulation with representative methods.
• Describe learned GDA techniques including graph structure learning, graph adversarial training, graph rationalization, and automated augmentation.
• Summarize GDA usage in self-supervised learning through contrastive, non-contrastive, and consistency objectives.
• Provide a consolidated view of challenges and future directions in Section 7.

실험 결과

연구 질문

• RQ1 What are the primary categories of graph data augmentation techniques and how can they be structured from multiple perspectives?
• RQ2 How do rule-based and learned GDA methods differ, and what are representative approaches in each category?
• RQ3 What are the current challenges and open directions for advancing graph data augmentation in GML?

주요 결과

• Rule-based GDA methods are the most commonly used due to simplicity and efficiency.
• GDA techniques are categorized into data removal, data addition, and data manipulation, and can be applied across node-, edge-, and graph-level tasks.
• Learned GDA approaches include Graph Structure Learning, Graph Adversarial Training, Graph Rationalization, and Automated Augmentation.
• GDA methods are also leveraged in self-supervised learning, using objectives such as contrastive, non-contrastive, and consistency training.
• The survey highlights unsolved challenges and future research directions, including domain adaptation, scalability, evaluation criteria, and theoretical foundations.