Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep learning for dynamic graphs: models and benchmarks

Alessio Gravina, Davide Bacciu

TL;DR

Dynamic graphs capture evolving relations in real systems, motivating a unified formalism for representation learning and a fair benchmarking framework. The authors survey static-graph DGNs, extend them to discrete- and continuous-time dynamic settings, and categorize methods into stacked, integrated, meta, autoencoder, random-walk, and hybrid architectures. They implement three benchmarks with standardized data, splits, and evaluation to fairly compare methods across tasks like node and link prediction, providing baseline performances and datasets for future research. The study underscores the importance of modeling temporal dynamics alongside topology, while also pointing to open issues such as heterogeneity, robustness, and long-range dependency preservation, with practical impact on real-world predictive systems.

Abstract

Recent progress in research on Deep Graph Networks (DGNs) has led to a maturation of the domain of learning on graphs. Despite the growth of this research field, there are still important challenges that are yet unsolved. Specifically, there is an urge of making DGNs suitable for predictive tasks on realworld systems of interconnected entities, which evolve over time. With the aim of fostering research in the domain of dynamic graphs, at first, we survey recent advantages in learning both temporal and spatial information, providing a comprehensive overview of the current state-of-the-art in the domain of representation learning for dynamic graphs. Secondly, we conduct a fair performance comparison among the most popular proposed approaches on node and edge-level tasks, leveraging rigorous model selection and assessment for all the methods, thus establishing a sound baseline for evaluating new architectures and approaches

Deep learning for dynamic graphs: models and benchmarks

TL;DR

Dynamic graphs capture evolving relations in real systems, motivating a unified formalism for representation learning and a fair benchmarking framework. The authors survey static-graph DGNs, extend them to discrete- and continuous-time dynamic settings, and categorize methods into stacked, integrated, meta, autoencoder, random-walk, and hybrid architectures. They implement three benchmarks with standardized data, splits, and evaluation to fairly compare methods across tasks like node and link prediction, providing baseline performances and datasets for future research. The study underscores the importance of modeling temporal dynamics alongside topology, while also pointing to open issues such as heterogeneity, robustness, and long-range dependency preservation, with practical impact on real-world predictive systems.

Abstract

Recent progress in research on Deep Graph Networks (DGNs) has led to a maturation of the domain of learning on graphs. Despite the growth of this research field, there are still important challenges that are yet unsolved. Specifically, there is an urge of making DGNs suitable for predictive tasks on realworld systems of interconnected entities, which evolve over time. With the aim of fostering research in the domain of dynamic graphs, at first, we survey recent advantages in learning both temporal and spatial information, providing a comprehensive overview of the current state-of-the-art in the domain of representation learning for dynamic graphs. Secondly, we conduct a fair performance comparison among the most popular proposed approaches on node and edge-level tasks, leveraging rigorous model selection and assessment for all the methods, thus establishing a sound baseline for evaluating new architectures and approaches
Paper Structure (31 sections, 34 equations, 4 figures, 20 tables)

This paper contains 31 sections, 34 equations, 4 figures, 20 tables.

Table of Contents

  1. Introduction
  2. Representation learning for static graphs
  3. Definitions and notations
  4. Overview of representation learning for static graphs
  5. Spectral convolution
  6. Spatial convolution
  7. Random walks
  8. Representation learning for dynamic Graphs: Notation and Taxonomy
  9. Learning with Discrete-Time Dynamic Graphs
  10. Spatio-temporal graphs
  11. Stacked architectures
  12. Integrated architectures
  13. General D-TDGs
  14. Integrated architectures
  15. Stacked architectures
  16. ...and 16 more sections

Figures (4)

  • Figure 1: Taxonomy employed to structure our survey of TDG models.
  • Figure 2: (a) A directed graph. (b) An undirected graph. (c) The neighborhood of node $u$.
  • Figure 3: Visual representation of a DGN. Given the input graph, each GCL $(\ell + 1)$ computes the new representation of a node $u$ as a transformation of $u$ and its neighbors representations at the previous layer, $\ell$.
  • Figure 4: (a) A Discrete-Time Dynamic Graph defined over five timestamps and a set of five interacting entities. (b) The evolution of a Continuous-Time Dynamic Graph through the stream of events until the timestamp $t_3$.