Attention Dispersion in Dynamic Graph Transformers: Diagnosis and a Transferable Fix

TL;DR

This paper diagnoses attention dispersion as a key failure in dynamic graph Transformers under temporal shift and proposes a simple, transferable fix called differential attention, leading to state-of-the-art results across multiple benchmarks.

Contribution

The paper identifies attention dispersion as a failure mode in CTDG Transformers under temporal shift and introduces differential attention as an effective, transferable solution.

Findings

Differential attention improves performance on high-shift datasets.

Attention entropy is reduced with differential attention.

DiffDyG achieves state-of-the-art results across 9 benchmarks.

Abstract

Transformer-based architectures have become the dominant paradigm for Continuous-Time Dynamic Graph (CTDG) learning, yet their performance remains limited on temporally shifted datasets. In this work, we identify attention dispersion as a shared failure mode of dynamic graph Transformers under temporal distribution shift. Through controlled ablation contrasting structurally and temporally distinguished historical neighbors against random ones, we show that prediction depends on a class of critical nodes that carry consistently more predictive signal than arbitrary neighbors. However, existing Transformers fail to focus on these nodes even when they are present in the input, as temporal shift weakens attention contrast and produces overly dispersed attention distributions. This diagnosis suggests a simple and transferable fix: replace standard attention with differential attention, which suppresses common-mode attention and amplifies distinctive token-level signals. When added to three representative CTDG Transformer baselines, differential attention consistently improves performance, with gains concentrated on high-shift datasets. Attention-level measurements further confirm the mechanism, showing reduced attention entropy and increased attention mass on critical nodes. Building on these findings, we introduce DiffDyG, a reference implementation combining differential attention with standard input encodings. Across 9 benchmarks and three negative sampling protocols, DiffDyG achieves SOTA performance, with especially large gains on the most shifted datasets.