TSGDiff: Rethinking Synthetic Time Series Generation from a Pure Graph Perspective

2025-11-15

AI Summary

This study addresses the challenge of modeling complex temporal dependencies and structural patterns in synthetic time series. We propose a purely graph-based generative framework: time series are mapped to topological graphs constructed from Fourier spectral features and dynamic time dependencies; a graph neural network encoder-decoder learns structure-aware latent representations; and a diffusion model enables high-fidelity generation. Key contributions include: (1) a dynamic graph construction method that explicitly captures joint frequency-domain and time-domain structural relationships; and (2) Topo-FID, a novel topology-aware fidelity metric integrating graph edit distance and structural entropy similarity to more accurately assess structural consistency of generated sequences. Extensive experiments on multiple real-world datasets demonstrate that our approach significantly improves preservation of temporal dependencies and structural integrity in generated series, achieving superior fidelity and practical utility over state-of-the-art baselines.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Natural Language Processing: Generation
• Computer Vision: Diffusion Models for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Diffusion models have shown great promise in data generation, yet generating time series data remains challenging due to the need to capture complex temporal dependencies and structural patterns. In this paper, we present extit{TSGDiff}, a novel framework that rethinks time series generation from a graph-based perspective. Specifically, we represent time series as dynamic graphs, where edges are constructed based on Fourier spectrum characteristics and temporal dependencies. A graph neural network-based encoder-decoder architecture is employed to construct a latent space, enabling the diffusion process to model the structural representation distribution of time series effectively. Furthermore, we propose the Topological Structure Fidelity (Topo-FID) score, a graph-aware metric for assessing the structural similarity of time series graph representations. Topo-FID integrates two sub-metrics: Graph Edit Similarity, which quantifies differences in adjacency matrices, and Structural Entropy Similarity, which evaluates the entropy of node degree distributions. This comprehensive metric provides a more accurate assessment of structural fidelity in generated time series. Experiments on real-world datasets demonstrate that extit{TSGDiff} generates high-quality synthetic time series data generation, faithfully preserving temporal dependencies and structural integrity, thereby advancing the field of synthetic time series generation.

Problem

• Generating time series data with complex temporal dependencies
• Modeling structural representation distribution via graph diffusion
• Assessing structural fidelity in synthetic time series generation

Innovation

• Representing time series as dynamic graphs
• Using graph neural network encoder-decoder architecture
• Proposing Topo-FID metric for structural similarity assessment