UTG: Towards a Unified View of Snapshot and Event Based Models for Temporal Graphs

TL;DR

The paper addresses the fragmentation between snapshot-based and event-based temporal graph learning by introducing Unified Temporal Graph (UTG), a framework with input/output mappers that lets models operate across CTDGs and DTDGs. It presents a UTG training procedure to boost snapshot-based models in streaming contexts and evaluates both representations on temporal link prediction, showing snapshot-based methods can approach event-based performance when enhanced by UTG while offering an order of magnitude faster inference. The findings highlight that joint neighborhood features drive strong performance across methods, regardless of data format, and they advocate combining the strengths of both paradigms to achieve scalable, high-performance temporal graph learning. This work provides a practical pathway to cross-validate architectures and encourages hybrid models that fuse efficiency with predictive power in real-time dynamic graphs.

Abstract

Many real world graphs are inherently dynamic, constantly evolving with node and edge additions. These graphs can be represented by temporal graphs, either through a stream of edge events or a sequence of graph snapshots. Until now, the development of machine learning methods for both types has occurred largely in isolation, resulting in limited experimental comparison and theoretical crosspollination between the two. In this paper, we introduce Unified Temporal Graph (UTG), a framework that unifies snapshot-based and event-based machine learning models under a single umbrella, enabling models developed for one representation to be applied effectively to datasets of the other. We also propose a novel UTG training procedure to boost the performance of snapshot-based models in the streaming setting. We comprehensively evaluate both snapshot and event-based models across both types of temporal graphs on the temporal link prediction task. Our main findings are threefold: first, when combined with UTG training, snapshot-based models can perform competitively with event-based models such as TGN and GraphMixer even on event datasets. Second, snapshot-based models are at least an order of magnitude faster than most event-based models during inference. Third, while event-based methods such as NAT and DyGFormer outperforms snapshot-based methods on both types of temporal graphs, this is because they leverage joint neighborhood structural features thus emphasizing the potential to incorporate these features into snapshotbased models as well. These findings highlight the importance of comparing model architectures independent of the data format and suggest the potential of combining the efficiency of snapshot-based models with the performance of event-based models in the future.

Figures (3)

• Figure 1: Illustration of the UTG framework. The input graph is processed by the UTG input mapper to generate the appropriate input data format for TG models. The model predictions are then processed by the UTG output mapper for prediction.
• Figure 2: Snapshot-based models with UTG training are at least an order of magnitude faster than event-based models for inference.
• Figure 3: Different setting for evaluation of future link prediction include between deployed, streaming and live-update setting. UTG framework is designed for the streaming setting.

Theorems & Definitions (8)

• Definition 1: Continuous Time Dynamic Graphs
• Definition 2: Discrete Time Dynamic Graphs
• Definition 3: Discretization Partition
• Definition 4: Regular Discretization Partition
• Definition 5: Induced Graph Snapshots
• Definition 6: Discretization Level
• Definition 7: Time Gap
• Definition 8: Zero-order Hold