Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

TiGer: Self-Supervised Purification for Time-evolving Graphs

Hyeonsoo Jo, Jongha Lee, Fanchen Bu, Kijung Shin

TL;DR

This work tackles noise in time-evolving graphs by introducing TiGer, a self-supervised purifier that explicitly models long-term contextual and short-term consistency patterns. It combines a self-attention-based long-term module with a statistical-distance-based short-term module through a proximity-aware ensemble, enabling robust edge purification without noise labels. Empirical results across five real-world datasets show TiGer achieves up to +10.2% purification accuracy improvements and up to +5.3% gains in downstream node classification, outperforming static-graph purifiers and several dynamic baselines. The approach is scalable, with a transparent training procedure and publicly available code to support reproducibility and practical deployment in dynamic graph analysis.

Abstract

Time-evolving graphs, such as social and citation networks, often contain noise that distorts structural and temporal patterns, adversely affecting downstream tasks, such as node classification. Existing purification methods focus on static graphs, limiting their ability to account for critical temporal dependencies in dynamic graphs. In this work, we propose TiGer (Time-evolving Graph purifier), a self-supervised method explicitly designed for time-evolving graphs. TiGer assigns two different sub-scores to edges using (1) self-attention for capturing long-term contextual patterns shaped by both adjacent and distant past events of varying significance and (2) statistical distance measures for detecting inconsistency over a short-term period. These sub-scores are used to identify and filter out suspicious (i.e., noise-like) edges through an ensemble strategy, ensuring robustness without requiring noise labels. Our experiments on five real-world datasets show TiGer filters out noise with up to 10.2% higher accuracy and improves node classification performance by up to 5.3%, compared to state-of-the-art methods.

TiGer: Self-Supervised Purification for Time-evolving Graphs

TL;DR

This work tackles noise in time-evolving graphs by introducing TiGer, a self-supervised purifier that explicitly models long-term contextual and short-term consistency patterns. It combines a self-attention-based long-term module with a statistical-distance-based short-term module through a proximity-aware ensemble, enabling robust edge purification without noise labels. Empirical results across five real-world datasets show TiGer achieves up to +10.2% purification accuracy improvements and up to +5.3% gains in downstream node classification, outperforming static-graph purifiers and several dynamic baselines. The approach is scalable, with a transparent training procedure and publicly available code to support reproducibility and practical deployment in dynamic graph analysis.

Abstract

Time-evolving graphs, such as social and citation networks, often contain noise that distorts structural and temporal patterns, adversely affecting downstream tasks, such as node classification. Existing purification methods focus on static graphs, limiting their ability to account for critical temporal dependencies in dynamic graphs. In this work, we propose TiGer (Time-evolving Graph purifier), a self-supervised method explicitly designed for time-evolving graphs. TiGer assigns two different sub-scores to edges using (1) self-attention for capturing long-term contextual patterns shaped by both adjacent and distant past events of varying significance and (2) statistical distance measures for detecting inconsistency over a short-term period. These sub-scores are used to identify and filter out suspicious (i.e., noise-like) edges through an ensemble strategy, ensuring robustness without requiring noise labels. Our experiments on five real-world datasets show TiGer filters out noise with up to 10.2% higher accuracy and improves node classification performance by up to 5.3%, compared to state-of-the-art methods.

Paper Structure

This paper contains 18 sections, 3 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Preliminaries and Related Work
  3. Background
  4. Graph Purification
  5. Temporal Patterns
  6. Proposed Method: TiGer
  7. Overview
  8. Long-term Module $M_{L}$
  9. Short-term Module $M_{S}$
  10. Ensemble Module
  11. Self-supervised Training Procedure
  12. Complexity Analysis
  13. Experiments
  14. Experiment Settings
  15. Q1. Accuracy
  16. ...and 3 more sections

Figures (3)

  • Figure 1: Overall process of graph purification using TiGer at each time step $t$.
  • Figure 2: (Q2) Effectiveness.TiGer consistently enhances node classification performance, outperforming its competitors in most cases. On the DBLP and ArXivAI datasets, GDC and Tiara exceed the six-hour time limit.
  • Figure : Consistency score