Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation

Ryien Hosseini, Filippo Simini, Venkatram Vishwanath, Henry Hoffmann

TL;DR

This paper tackles the challenge of generating continuously evolving graphs by CTDGs. It introduces DG-Gen, an encoder–decoder framework that directly models the joint probability of temporal interactions as $p(src)\,p(dst|src)\,p(t, \mathbf{e}|dst, src)$, enabling autoregressive generation without recourse to static snapshots. The method leverages a TGN-based encoder and a deep probabilistic decoder with modular components (Reshape, Product, Merge, Time+MSG) to produce both event times and edge features, achieving inductive generation that handles arbitrarily-sized features and unseen nodes. Empirically, DG-Gen outperforms the inductive baseline TIGGER-I across five datasets in graph fidelity and achieves superior link-prediction performance, demonstrating the practical impact of direct temporal-event modeling for scalable, realistic CTDG generation.

Abstract

Recent advancements in graph representation learning have shifted attention towards dynamic graphs, which exhibit evolving topologies and features over time. The increased use of such graphs creates a paramount need for generative models suitable for applications such as data augmentation, obfuscation, and anomaly detection. However, there are few generative techniques that handle continuously changing temporal graph data; existing work largely relies on augmenting static graphs with additional temporal information to model dynamic interactions between nodes. In this work, we propose a fundamentally different approach: We instead directly model interactions as a joint probability of an edge forming between two nodes at a given time. This allows us to autoregressively generate new synthetic dynamic graphs in a largely assumption free, scalable, and inductive manner. We formalize this approach as DG-Gen, a generative framework for continuous time dynamic graphs, and demonstrate its effectiveness over five datasets. Our experiments demonstrate that DG-Gen not only generates higher fidelity graphs compared to traditional methods but also significantly advances link prediction tasks.

A Deep Probabilistic Framework for Continuous Time Dynamic Graph Generation

TL;DR

This paper tackles the challenge of generating continuously evolving graphs by CTDGs. It introduces DG-Gen, an encoder–decoder framework that directly models the joint probability of temporal interactions as , enabling autoregressive generation without recourse to static snapshots. The method leverages a TGN-based encoder and a deep probabilistic decoder with modular components (Reshape, Product, Merge, Time+MSG) to produce both event times and edge features, achieving inductive generation that handles arbitrarily-sized features and unseen nodes. Empirically, DG-Gen outperforms the inductive baseline TIGGER-I across five datasets in graph fidelity and achieves superior link-prediction performance, demonstrating the practical impact of direct temporal-event modeling for scalable, realistic CTDG generation.

Abstract

Recent advancements in graph representation learning have shifted attention towards dynamic graphs, which exhibit evolving topologies and features over time. The increased use of such graphs creates a paramount need for generative models suitable for applications such as data augmentation, obfuscation, and anomaly detection. However, there are few generative techniques that handle continuously changing temporal graph data; existing work largely relies on augmenting static graphs with additional temporal information to model dynamic interactions between nodes. In this work, we propose a fundamentally different approach: We instead directly model interactions as a joint probability of an edge forming between two nodes at a given time. This allows us to autoregressively generate new synthetic dynamic graphs in a largely assumption free, scalable, and inductive manner. We formalize this approach as DG-Gen, a generative framework for continuous time dynamic graphs, and demonstrate its effectiveness over five datasets. Our experiments demonstrate that DG-Gen not only generates higher fidelity graphs compared to traditional methods but also significantly advances link prediction tasks.

Paper Structure

This paper contains 22 sections, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Problem Formulation.
  3. Limitations of Prior Approaches.
  4. Our Key Insight.
  5. Our Solution.
  6. Contributions.
  7. Background and Related Work
  8. Dynamic Graph Representations
  9. Discriminative Models for CTDGs
  10. Generative Models for Dynamic Graph Data
  11. Generating DTDGs.
  12. Generating CTDGs.
  13. Inductive versus Transductive Graph Generation.
  14. Advancing CTDG Generation.
  15. The Dynamic Graph Generative Network
  16. ...and 7 more sections

Figures (3)

  • Figure 1: Overview of DG-Gen's architecture and internal modules, described in Section \ref{['sec:model']}.
  • Figure 2: Top panels: Histograms of one randomly selected edge feature distribution per dataset of the real data (blue) and of the synthetic data generated by DG-Gen (orange). Bottom panels: Jensen-Shannon distances between the 2-dimensional histograms (real and synthetic) of the joint distributions of feature pairs.
  • Figure 3: Average Precision for DG-Gen and baselines on link prediction via inductive sampling.