Efficient Learning-based Graph Simulation for Temporal Graphs

TL;DR

This paper addresses the challenge of efficiently generating temporal graphs that preserve both structural and temporal properties observed in real data. It introduces Temporal Graph Autoencoder (TGAE), a four-part framework that uses k-radius temporal ego-graphs and TGAT-based encoding to learn local temporal structure, followed by an ego-graph decoder and a scalable assembling process. The authors propose GPU-friendly parallel training and several model variants to balance quality and efficiency, achieving superior results in temporal motif preservation and overall generation quality while reducing training and inference costs. Empirical results on seven real temporal networks demonstrate that TGAE outperforms state-of-the-art temporal graph generators in both fidelity and scalability, with practical implications for large-scale dynamic graph synthesis and simulation.

Abstract

Graph simulation has recently received a surge of attention in graph processing and analytics. In real-life applications, e.g. social science, biology, and chemistry, many graphs are composed of a series of evolving graphs (i.e., temporal graphs). While most of the existing graph generators focus on static graphs, the temporal information of the graphs is ignored. In this paper, we focus on simulating temporal graphs, which aim to reproduce the structural and temporal properties of the observed real-life temporal graphs. In this paper, we first give an overview of the existing temporal graph generators, including recently emerged learning-based approaches. Most of these learning-based methods suffer from one of the limitations: low efficiency in training or slow generating, especially for temporal random walk-based methods. Therefore, we propose an efficient learning-based approach to generate graph snapshots, namely temporal graph autoencoder (TGAE). Specifically, we propose an attention-based graph encoder to encode temporal and structural characteristics on sampled ego-graphs. And we proposed an ego-graph decoder that can achieve a good trade-off between simulation quality and efficiency in temporal graph generation. Finally, the experimental evaluation is conducted among our proposed TGAE and representative temporal graph generators on real-life temporal graphs and synthesized graphs. It is reported that our proposed approach outperforms the state-of-the-art temporal graph generators by means of simulation quality and efficiency.

Figures (6)

• Figure 1: An example of time-evolving graph.
• Figure 2: The framework of our proposed TGAE. TGAE is composed of four parts: (1) ego-graph sampling, (2) temporal graph encoding, (3) ego-graph decoding, and (4) temporal graph assembling.
• Figure 3: The illustration of the $k$-radius temporal ego-graph. The upper left part shows the ego-graph with the center temporal node $u^t$. The other three parts shows the edge importances calculated by $k$ stacked temporal graph attention (TGAT) layers.
• Figure 4: The illustration of the $k$-bipartite computation graphs. The upper part shows the initial $k$-radius temporal ego-graphs. The lower part shows the $k$-bipartite computation graphs, which are used for model training. In each bipartite computation graph, the results of the target nodes can be computed concurrently.
• Figure 5: The comparison results on the seven evaluation metrics across 15 timestamps in DBLP data set. Best viewed in color. The algorithm better fitting the curve of the original graph (colored in blue) is better.

Theorems & Definitions (4)

• Definition 1
• Definition 2
• Definition 3
• Definition 4