Generating Causal Temporal Interaction Graphs for Counterfactual Validation of Temporal Link Prediction

Temporal link prediction (TLP) models are commonly evaluated based on predictive accuracy, yet such evaluations do not assess whether these models capture the causal mechanisms that govern temporal interactions. In this work, we propose a framework for counterfactual validation of TLP models by generating causal temporal interaction graphs (CTIGs) with known ground-truth causal structure. We first introduce a structural equation model for continuous-time event sequences that supports both excitatory and inhibitory effects, and then extend this mechanism to temporal interaction graphs. To compare causal models, we propose a distance metric based on cross-model predictive error, and empirically validate the hypothesis that predictors trained on one causal model degrade when evaluated on sufficiently distant models. Finally, we instantiate counterfactual evaluation under (i) controlled causal shifts between generating models and (ii) timestamp shuffling as a stochastic distortion with measurable causal distance. Our framework provides a foundation for causality-aware benchmarking.

💡 Research Summary

The paper addresses a critical gap in temporal link prediction (TLP) research: current evaluations focus solely on predictive accuracy and ignore whether models capture the underlying causal mechanisms that generate temporal interactions. To remedy this, the authors propose a comprehensive framework for counterfactual validation based on Causal Temporal Interaction Graphs (CTIGs), which are synthetic datasets with fully known causal structure.

First, they introduce a continuous‑time structural equation model (SEM) for event sequences. Each event type i has a trigger process Uᵢ(t) (a homogeneous Poisson stream) and a set of past‑window indicators X′ⱼ(t) that record whether parent events j occurred within a delay window (\bar τ). The occurrence of i at time t is given by
(X_i(t)=U_i(t)·\mathbf{I}{\sum_{j∈P_i} Θ_{i,j} X′_j(t) ≥ 0}),
where the causal influence parameters Θᵢⱼ∈