Event-Order Constraints for Anchor-Scarce Alignment of Evolving Knowledge Graphs

Abstract

The alignment of knowledge graphs is a foundational task in data integration, enabling the synthesis of fragmented information across diverse domains. However, traditional alignment methodologies rely heavily on the availability of predefined anchor links and assume that the underlying graphs are static. In real-world scenarios, knowledge graphs are continuously evolving, characterized by the frequent addition, deletion, and modification of facts over time. Furthermore, due to privacy concerns, structural heterogeneity, and varying naming conventions, establishing sufficient anchor links between different knowledge graphs is often impractical, leading to an anchor-scarce environment. This paper introduces a novel framework that leverages event-order constraints to facilitate robust entity alignment in evolving knowledge graphs under severe anchor scarcity. By conceptualizing temporal facts as ordered events, the proposed framework constructs invariant temporal signatures for entities. These signatures capture the inherent logical and chronological dependencies between events, which remain consistent across different knowledge graph representations even when superficial entity attributes or structural topologies diverge. Through comprehensive theoretical analysis and extensive empirical evaluations across multiple benchmark datasets, we demonstrate that incorporating event-order constraints significantly improves alignment accuracy compared to state-of-the-art structural and temporal alignment models. The results indicate that temporal logic provides a highly effective surrogate for missing topological anchors, offering a new pathway for integrating dynamic, heterogeneous data systems.