Towards the Explainability of Temporal Graph Networks via Memory Backtracking and Topological Attribution

2026-07-04

AI Summary

Existing explanation methods for temporal graph networks (TGNs) struggle to capture the influence of historical events on predictions, particularly overlooking the role of memory modules. This work addresses this gap by introducing, for the first time, a dual-tree explanation framework that explicitly incorporates memory mechanisms: a topological attribution tree and a memory backtracking tree jointly model topological and temporal dependencies. By integrating Layer-wise Relevance Propagation (LRP) with a tailored optimization objective, the method quantifies contributions from neighboring nodes and past events, while an improved top-k selection strategy mitigates faithfulness issues caused by nonlinear mappings. Extensive experiments across nine temporal graph datasets demonstrate that the proposed approach significantly outperforms existing baselines in node attribute prediction, link prediction, and graph classification tasks, achieving markedly higher explanation fidelity.

Abstract

Temporal graphs are ubiquitous in real-world applications and Temporal Graph Networks (TGNs) have achieved superior predictive accuracy. Understanding which historical events drive model predictions can enhance trustworthiness of TGNs. Existing explanation methods overlook the memory module, the core component that records and updates node histories, leaving the influence of past events unexplored. To address this, we attribute TGNs predictions through the topology attribution tree and memory backtracking tree. The topology attribution tree captures the influence of neighbors and their memory vectors, then the memory backtracking tree quantifies how historical events shape node memory vectors. We apply the LRP in TGNs, ensuring that the total contribution of events equals the logits of model. Finally, top-k selection may be unfaithful due to the nonlinear mapping from logits to probabilities, we design optimization objectives to identify the important events. Experiments on nine temporal graph datasets, spanning node property prediction, link prediction tasks and graph classification tasks, show that our method provides faithful explanations and outperforms state-of-the-art baselines. The code is available at https://github.com/yazhengliu/MemExplainer

Problem

• Temporal Graph Networks
• Explainability
• Memory Module
• Historical Events
• Model Interpretability

Innovation

• Temporal Graph Networks
• Memory Backtracking
• Topological Attribution
• Explainability
• Layer-wise Relevance Propagation