Graph Neural Networks for temporal graphs - State of the art, open challenges, and opportunities

Aug 5, 2026

Graph Neural Networks (GNNs) are widely used for learning on static graph-structured data, yet many real-world systems evolve as graphs and attributes change over time. This work surveys Temporal Graph Neural Networks by formally defining learning settings and tasks, then introduces a taxonomy that categorizes existing methods by how temporal information is represented and processed. The study concludes with key open challenges and discusses research and application perspectives for future progress.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages21

Reading Time53 min

AccessFree to download

Introduction

Temporal graphs in real-world applications

Temporal Graphs

Formal definitions: static vs temporal graphs

Time-dependent nodes and edges

arXiv :2302 .0 10 18v4 [ cs .LG] 8 Jul 2023 Graph Neural Networks for temporal graphs: State of the art, open challenges, and opportunities Antonio Longa 1,2,* , Veronica Lachi3,* , Gabriele Santin 1,* , Monica Bianchini3 , Bruno Lepri 1 , Pietro Li`o4 , Franco Scarselli3 , and Andrea Passerini2 1 Fondazione Bruno Kessler, Trento, Italy 2 University of Trento, Trento, Italy 3 University of Siena, Siena, Italy 4 University of Cambridge, Cambridge, United Kingdom * Equal contribution July 11, 2023 Abstract Graph Neural Networks (GNNs) have become the leading paradigm for learning on (static) graphstructured data. However, many real-world systems are dynamic in nature, since the graph and node/edge attributes change over time. In recent years, GNNbased models for temporal graphs have emerged as a promising area of research to extend the capabilities of GNNs. In this work, we provide the first comprehensive overview of the current state-of-the-art of temporal GNN, introducing a rigorous formalization of learning settings and tasks and a novel taxonomy categorizing existing approaches in terms of how the temporal aspect is represented and processed. We conclude the survey with a discussion of the most relevant open challenges for the field, from both research and application perspectives. 1 Introduction The ability to process temporal graphs is becoming increasingly important in a variety of fields such as recommendation systems [20, 88], social network analysis [15, 19], transportation systems [36, 101], face-to-face interactions [48], human mobility [52, 24], epidemic modeling and contact tracing [8, 74], and many others. Traditional graph-based models are not well suited for analyzing temporal graphs as they assume a fixed structure and are unable to capture its temporal evolution. Therefore, in the last few years, several models capable to directly encode temporal graphs have been developed, such as matrix factorization-based approaches [1] and temporal motif-based methods [47] . Recently, also GNNs have been successfully applied to temporal graphs. Indeed, their success in various static graph tasks, including node classification [30, 81, 39, 22, 25, 56, 87] and link prediction [102, 7, 103], has not only established them as the leading paradigm in static graph processing, but has also indicated the importance of exploring their potential in other graph domains, such as temporal graphs. With approaches ranging from attention-based methods [92] to Variational GraphAutoencoders (VGAEs) [29], Temporal Graph Neural Networks (TGNNs) have achieved state-of-the-art results on tasks such as temporal link prediction [70], node classification [66] and edge classification [83] . Despite the potential of GNN-based models for tem- poral graph processing and the variety of different approaches that emerged, a systematization of the literature is still missing. Existing surveys either discuss general techniques for learning over temporal graphs, only briefly mentioning temporal extensions of GNNs [37, 3, 96, 91], or focus on specific topics, like temporal link prediction [63, 73] or temporal graph generation [28], or present an overview of GNN models designed for different types of graphs without providing in-depth coverage of temporal GNNs [77] . This work aims to fill this gap by providing a systematization of existing GNN-based methods for temporal graphs and a formalization of the tasks being addressed. Our main contributions are the following: • We propose a coherent formalization of the different learning settings and of the tasks that can be performed on temporal graphs, unifying existing formalism and informal definitions that are scattered in the literature, and highlighting substantial gaps in what is currently being tackled; • We organize existing TGNN works into a comprehensive taxonomy that groups methods according to the way in which time is represented and the mechanism with which it is ta

FAQs

Why are static GNN models insufficient for temporal graphs?

Static graph models assume a fixed structure and cannot represent how edges, nodes, and attributes evolve over time, which is central to temporal graph problems.

What does the paper contribute for Temporal Graph Neural Networks?

It provides a rigorous formalization of learning settings and tasks, and a taxonomy organizing temporal GNN methods based on how time is represented and processed.

What is discussed at the end of the survey?

The survey highlights the most relevant open challenges from both research and application perspectives, outlining directions for further investigation.

Aug 5, 2026

Upload by Genevieve