Future Link Prediction Without Memory or Aggregation

TL;DR

This work tackles future link prediction on temporal graphs, addressing the challenge of predicting both seen and unseen edges. It introduces CRAFT, a memory-free architecture that replaces memory and aggregation modules with learnable node embeddings and cross-attention between candidate destinations and the source's recent neighbors to enable target-aware matching. The approach yields strong improvements across 17 datasets, particularly in unseen-edge scenarios, while maintaining high efficiency suitable for large-scale deployment. The study demonstrates that explicit node identifiers and destination-context matching are key to robust temporal link prediction in real-world dynamic systems.

Abstract

Future link prediction on temporal graphs is a fundamental task with wide applicability in real-world dynamic systems. These scenarios often involve both recurring (seen) and novel (unseen) interactions, requiring models to generalize effectively across both types of edges. However, existing methods typically rely on complex memory and aggregation modules, yet struggle to handle unseen edges. In this paper, we revisit the architecture of existing temporal graph models and identify two essential but overlooked modeling requirements for future link prediction: representing nodes with unique identifiers and performing target-aware matching between source and destination nodes. To this end, we propose Cross-Attention based Future Link Predictor on Temporal Graphs (CRAFT), a simple yet effective architecture that discards memory and aggregation modules and instead builds on two components: learnable node embeddings and cross-attention between the destination and the source's recent interactions. This design provides strong expressive power and enables target-aware modeling of the compatibility between candidate destinations and the source's interaction patterns. Extensive experiments on diverse datasets demonstrate that CRAFT consistently achieves superior performance with high efficiency, making it well-suited for large-scale real-world applications.

Figures (11)

• Figure 1: The general model architecture of existing temporal graph learning methods.
• Figure 2: Model architecture of our proposed method, CRAFT: representing nodes with learnable embeddings and performing target-aware matching using cross-attention.
• Figure 3: Illustration of different source-destination matching schemes for predicting whether UserA is likely to purchase iPhone Case at time $t$. Before time $t$, UserA has purchased iPhone, AirPods, and iPad in chronological order, while UserB and UserC have previously purchased iPhone Case.
• Figure 4: Test set inference time of CRAFT and baselines on Flights, Yelp and WikiLink.
• Figure 5: Evaluating the impact of learnable node embeddings and cross-attention: CRAFT vs. TGAT/DyGFormer with and without learnable embeddings (LE).