Brief

Researchers have introduced TGFormer, a new Transformer architecture designed to improve the modeling of temporal graphs. This model addresses limitations in capturing long-term dependencies and identifying periodic patterns within these dynamic networks. By employing a trajectory framework and an auto-correlation mechanism, TGFormer systematically analyzes historical interactions to derive node representations and uncover periodic dependencies, leading to significant efficiency and accuracy gains. AI

Researchers have introduced ST-TGExplainer, a novel method designed to improve the interpretability of Temporal Graph Neural Networks (TGNNs). Existing models often struggle to distinguish between the influence of past interactions (stability patterns) and newly emerging ones (transition patterns) on predictions. ST-TGExplainer addresses this by learning a compact explanatory subgraph that disentangles these two pattern types, ensuring predictive accuracy while reducing redundancy. Experiments show that this approach leads to more faithful explanations for TGNN predictions. AI