[Paper Review] Data Augmentation for Deep Graph Learning: A Survey

Kaize Ding, Zhe Xu+2 more|arXiv (Cornell University)|Feb 16, 2022

Advanced Graph Neural Networks27 citations

TL;DR

This survey formalizes graph data augmentation (GraphDA), provides a taxonomy by augmentation modality, and reviews GraphDA techniques and their applications to low-resource and reliable graph learning.

ABSTRACT

Graph neural networks, a powerful deep learning tool to model graph-structured data, have demonstrated remarkable performance on numerous graph learning tasks. To address the data noise and data scarcity issues in deep graph learning, the research on graph data augmentation has intensified lately. However, conventional data augmentation methods can hardly handle graph-structured data which is defined in non-Euclidean space with multi-modality. In this survey, we formally formulate the problem of graph data augmentation and further review the representative techniques and their applications in different deep graph learning problems. Specifically, we first propose a taxonomy for graph data augmentation techniques and then provide a structured review by categorizing the related work based on the augmented information modalities. Moreover, we summarize the applications of graph data augmentation in two representative problems in data-centric deep graph learning: (1) reliable graph learning which focuses on enhancing the utility of input graph as well as the model capacity via graph data augmentation; and (2) low-resource graph learning which targets on enlarging the labeled training data scale through graph data augmentation. For each problem, we also provide a hierarchical problem taxonomy and review the existing literature related to graph data augmentation. Finally, we point out promising research directions and the challenges in future research.

Motivation & Objective

• Formally formulate the GraphDA problem and its challenges.
• Propose a comprehensive taxonomy of GraphDA techniques by augmentation modality.
• Review GraphDA applications in low-resource graph learning and reliable graph learning.
• Summarize training paradigms (decoupled, joint, bi-level) for learnable GraphDA methods.
• Highlight open issues and promising directions for future GraphDA research.

Proposed method

• Define augmentation transformation f_theta(G) to produce tilde_G with augmented adjacency and feature matrices.
• Classify augmentation methods as structure-, feature-, or label-oriented.
• Describe three training paradigms for learnable GraphDA: decoupled training, joint training, and bi-level optimization.
• Introduce augmentation and utility losses L_aug and L_utility and their roles in optimization.
• Provide a structured literature review of GraphDA techniques and their applications to different DGL tasks.

Experimental results

Research questions

• RQ1 What is graph data augmentation (GraphDA) and how can it be formally defined?
• RQ2 How can GraphDA techniques be systematically categorized by augmentation modality?
• RQ3 How are GraphDA techniques applied to low-resource graph learning and reliable graph learning?
• RQ4 What training strategies and objectives best integrate GraphDA with downstream GNN tasks?

Key findings

• This is the first survey dedicated to GraphDA, providing formal formulation and a unified review.
• A comprehensive taxonomy of GraphDA techniques by target modality (structure-, feature-, label-oriented) is presented.
• Applications of GraphDA to low-resource graph learning (SSL and semi-supervised) and reliable graph learning are analyzed.
• The paper discusses training paradigms (decoupled, joint, bi-level) and outlines open challenges and future directions.
• It connects GraphDA techniques to practical graph tasks and highlights design principles for effective augmentation.