TaCE: Time-aware Convolutional Embedding Learning for Temporal Knowledge Graph Completion
Temporal knowledge graph completion (TKGC) is a challenging task to infer the missing component for quadruples. The key challenge lies at how to integrate time information into the embeddings of entities and relations. Recent TKGC methods tend to capture temporal patterns via linear or multilinear models, which are fast but not expressive enough. In this study, we propose a novel time-aware convolutional embedding model (TaCE) to represent the time-dependent facts in the task of TKGC. It highlights its novelty to feasibly convert timestamps as temporal convolutional filters to fully interact with entities and relations and learn temporal patterns in knowledge graphs (KGs). An extensive comparison proves that our model outperforms the state-of-the-art models on three public benchmark datasets of ICEWS14, ICEWS05-15 and GDELT. Results also demonstrate good temporal expressiveness and computation efficiency performed by our TaCE.
PDF Abstract