GAttNHP: Group Attention Neural Hawkes Process for Extrapolation Reasoning in Temporal Knowledge Graphs

2026-07-16

AI Summary

This work addresses three key challenges in future event prediction on temporal knowledge graphs: difficulty in modeling long-range dependencies, complex excitation relationships across event sequences, and the unreliability of deterministic predictions under heavy-tailed inter-event time distributions. To tackle these issues, the authors propose GAttNHP, a novel framework that uniquely integrates grouped attention mechanisms with neural Hawkes processes. GAttNHP employs a self-attention encoder to capture distant historical influences and introduces a semantic soft grouping module to enable efficient excitation sharing across event chains. Furthermore, it incorporates a learnable global prior to generate cross-attention masks, reducing computational complexity, and adopts a non-crossing quantile regression head to enhance robustness to heavy-tailed temporal distributions. Experiments on six benchmark datasets demonstrate that GAttNHP consistently outperforms state-of-the-art methods in both entity and timestamp prediction, with particularly notable gains on long-tailed event sequences.

Abstract

Temporal Knowledge Graphs (TKGs) record how facts evolve over time, but forecasting future events on a TKG remains difficult for three reasons: (i) long-range temporal dependencies are hard to encode; (ii) events on different chains mutually excite or inhibit one another in ways that snapshot-level models cannot express; and (iii) inter-arrival times are heavy-tailed and statistically sparse, so deterministic time predictors are unreliable. We address these three issues with a single framework, the \textbf{Group Attention Neural Hawkes Process (GAttNHP)}, built around three matched components. First, a self-attention encoder casts each subject--relation chain as a continuous-time point process and captures the lingering excitation of distant history. Second, a semantic soft-grouping module turns globally learnable Hawkes priors into an analytical cross-attention mask, so chains share excitation patterns through their latent group memberships rather than through exhaustive pairwise computation. Third, a Non-Crossing Quantile (NCQ) regression head replaces mean-based time prediction, providing calibrated, monotonically ordered quantile estimates that remain stable under heavy-tailed inter-arrival distributions. On six benchmark TKG datasets, GAttNHP improves over state-of-the-art baselines on both entity prediction and time prediction, and ablations confirm that its largest gains arise on the long-tail event chains where existing models fail most severely.

Problem

• Temporal Knowledge Graphs
• long-range temporal dependencies
• mutual excitation and inhibition
• heavy-tailed inter-arrival times
• event forecasting

Innovation

• Group Attention
• Neural Hawkes Process
• Temporal Knowledge Graphs
• Non-Crossing Quantile Regression
• Soft Grouping