[Paper Review] Graph-based Semi-supervised Learning: A Comprehensive Review

Zixing Song, Xiangli Yang+2 more|arXiv (Cornell University)|Feb 26, 2021

Text and Document Classification Technologies26 citations

TL;DR

This survey provides a systematic, up-to-date taxonomy and critical review of graph-based semi-supervised learning (GSSL), detailing graph construction and label-inference methods, with a unified encoder–decoder perspective for graph embedding and discussions of future directions.

ABSTRACT

Semi-supervised learning (SSL) has tremendous value in practice due to its ability to utilize both labeled data and unlabelled data. An important class of SSL methods is to naturally represent data as graphs such that the label information of unlabelled samples can be inferred from the graphs, which corresponds to graph-based semi-supervised learning (GSSL) methods. GSSL methods have demonstrated their advantages in various domains due to their uniqueness of structure, the universality of applications, and their scalability to large scale data. Focusing on this class of methods, this work aims to provide both researchers and practitioners with a solid and systematic understanding of relevant advances as well as the underlying connections among them. This makes our paper distinct from recent surveys that cover an overall picture of SSL methods while neglecting fundamental understanding of GSSL methods. In particular, a major contribution of this paper lies in a new generalized taxonomy for GSSL, including graph regularization and graph embedding methods, with the most up-to-date references and useful resources such as codes, datasets, and applications. Furthermore, we present several potential research directions as future work with insights into this rapidly growing field.

Motivation & Objective

• Provide a solid, systematic understanding of GSSL advances and how they relate.
• Propose a generalized taxonomy separating graph construction from label inference.
• Summarize graph regularization and graph embedding approaches within a unified framework.
• Offer practical resources including codes, datasets, and applications for researchers and practitioners.
• Identify open problems and future directions in dynamicity, scalability, noise resilience, and robustness.

Proposed method

• Introduce a two-step GSSL process: graph construction and label inference.
• Present a generalized loss framework L(f) = Ls(f, Dl) + μLr(f, D) and specialize it to graph regularization and graph embedding.
• Classify label inference methods into graph regularization and graph embedding; discuss specific regularizers and their formulations.
• Propose a unified encoder–decoder view for graph embeddings to connect shallow and deep embedding approaches.
• Review specific methods: label propagation, GRF, LGC, p-Laplacian regularization, directed regularization, manifold regularization, LPDGL, Poisson learning.
• Discuss graph construction techniques such as KNN, ε-neighborhood, and b-matching, including supervised variants.

Experimental results

Research questions

• RQ1 How can graph construction and label inference be systematically categorized in GSSL?
• RQ2 What are the core regularization and embedding approaches that enable effective label propagation on graphs?
• RQ3 How can graph embeddings be unified under an encoder–decoder framework to capture both shallow and deep representations?
• RQ4 What resources (codes, datasets, applications) support practical adoption and experimentation in GSSL?
• RQ5 What are the open research directions for dynamicity, scalability, noise resilience, and robustness in GSSL?

Key findings

• The paper provides a comprehensive, up-to-date review and a new taxonomy for GSSL.
• It presents a general loss framework guiding both graph regularization and graph embedding methods.
• Label propagation and its variants (GRF, LGC, p-Laplacian, directed regularization) are central to graph regularization.
• Graph embedding is framed within an encoder–decoder paradigm, unifying shallow and deep approaches.
• The survey collates available codes, datasets, and applications to support practical experimentation.
• It discusses open problems and future directions in dynamic, scalable, and robust GSSL.