TiGer: Self-Supervised Purification for Time-evolving Graphs

2025-03-10

AI Summary

Temporal graphs—such as social and citation networks—are often corrupted by noise, degrading structural integrity, temporal patterns, and downstream task performance. Existing denoising methods primarily target static graphs and lack effective modeling of temporal dependencies in dynamic settings. To address this, we propose TiGer, a self-supervised framework for dynamic graph denoising. TiGer is the first to introduce self-supervised learning for noise identification in temporal graphs, jointly capturing long-range temporal dependencies via multi-scale self-attention and short-term statistical anomalies using statistical distances (e.g., Jensen–Shannon divergence). It further employs an edge-level dual-scoring mechanism and a weighted ensemble filtering strategy to enhance robustness. Evaluated on five real-world datasets, TiGer achieves up to 10.2% higher noise identification accuracy and improves node classification performance by up to 5.3%, significantly outperforming state-of-the-art baselines.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Planning, Routing, and Scheduling: Temporal Planning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• Purifies noise in time-evolving graphs like social networks.
• Addresses limitations of static graph purification methods.
• Improves node classification by filtering suspicious edges.

Innovation

• Self-supervised purification for time-evolving graphs
• Uses self-attention for long-term contextual patterns
• Employs statistical distance for short-term inconsistency detection