🤖 AI Summary
Existing temporal knowledge graph (TKG) reasoning methods are vulnerable to spurious correlations induced by confounding features, leading to poor robustness in link prediction due to reliance on non-causal associations rather than underlying causal mechanisms. This paper introduces, for the first time, causal intervention theory into temporal graph representation learning, proposing a causal disentanglement framework that explicitly separates causal representations (entity- and relation-specific) from confounded representations. By performing counterfactual interventions, the framework identifies and suppresses confounding variables, enabling causal representations to drive link prediction. The method integrates temporal graph neural networks, causal structure learning, and confounder suppression. Evaluated on six standard benchmarks, it consistently outperforms state-of-the-art approaches in both predictive accuracy and robustness, demonstrating that causal enhancement fundamentally improves temporal knowledge reasoning.
📝 Abstract
Temporal knowledge graph reasoning (TKGR) is increasingly gaining attention for its ability to extrapolate new events from historical data, thereby enriching the inherently incomplete temporal knowledge graphs. Existing graph-based representation learning frameworks have made significant strides in developing evolving representations for both entities and relational embeddings. Despite these achievements, there's a notable tendency in these models to inadvertently learn biased data representations and mine spurious correlations, consequently failing to discern the causal relationships between events. This often leads to incorrect predictions based on these false correlations. To address this, we propose an innovative Causal Enhanced Graph Representation Learning framework for TKGR (named CEGRL-TKGR). This framework introduces causal structures in graph-based representation learning to unveil the essential causal relationships between events, ultimately enhancing the performance of the TKGR task. Specifically, we first disentangle the evolutionary representations of entities and relations in a temporal knowledge graph sequence into two distinct components, namely causal representations and confounding representations. Then, drawing on causal intervention theory, we advocate the utilization of causal representations for predictions, aiming to mitigate the effects of erroneous correlations caused by confounding features, thus achieving more robust and accurate predictions. Finally, extensive experimental results on six benchmark datasets demonstrate the superior performance of our model in the link prediction task.