Temporal Correlation Embedding for Enhanced Temporal Knowledge Graph Evolution

Temporal knowledge graphs (TKGs) extend traditional knowledge graphs by incorporating temporal information, enabling the modeling of dynamic facts. However, they often suffer from incompleteness and are poorly handled by traditional models, which typically treat time as a scalar or fixed vector, ignoring dependencies between adjacent timestamps. This limits their ability to capture temporal dynamics and relationship evolution. To address these issues, we propose the Temporal Correlation Embedding (TCE) model. TCE leverages GRU-based dynamic relation embeddings to capture long-term dependencies and the evolution of entity relations over time. It enhances temporal representations by embedding correlations between adjacent timestamps and learns how relationships change over time. Experiments on benchmark datasets show that TCE outperforms existing models in representing evolving relationships in TKGs.