QLGAN: a quantum-lineage graph attention network for temporal knowledge graph entity alignment

Temporal entity alignment, the task of identifying equivalent entities across evolving knowledge graphs (KGs), is a critical yet challenging problem. Existing methods often struggle to holistically model the complex interplay between structural topology and temporal dynamics while also failing to capture long-range dependencies encoded in multi-hop relational paths. To address these limitations, we propose the Quantum-Lineage Graph Attention Network (QLGAN), a novel quantum-classical hybrid model for temporal entity alignment. QLGAN uniquely integrates a path encoder directly within its attention mechanism, which enriches neighbor representations with multi-hop relational semantics before aggregation. Furthermore, it leverages a Variational Quantum Circuit (VQC) to perform powerful non-linear feature transformations, enabling a more effective fusion of complex spatiotemporal features. The model’s alignment process is based on a hybrid similarity framework that combines quantum-inspired metrics with structurally aware temporal propagation, followed by an optimal transport alignment algorithm. Extensive experiments on benchmark datasets validate the effectiveness of our approach in learning robust and contextually-rich representations for temporal knowledge graphs.