TFOLQ: first-order logic query over temporal knowledge graph with neural logical operators and hybrid query execution

Temporal Knowledge Graphs (TKGs) extend the temporal dimension on the basis of knowledge graph structure and model relevant facts that evolve over time. The existing static knowledge graph query methods cannot effectively adapt to the intricate logic composition required by complex queries, and struggles to handle diversified logic combination queries and multi-hop path queries. To address these challenges, we propose the Temporal First-Order Logic Query (TFOLQ) framework, a differentiable framework for complex queries on temporal knowledge graphs. TFOLQ extends the pre-trained temporal embedding model by merging learnable neural logical operators (projection, intersection, union, and negation) in fractional space, thus enabling diverse logical combination queries. Furthermore, we propose a hybrid query execution mechanism that fuses rule-guided path queries with neural logic queries to address multi-hop path queries. Extensive experiments with ICEWS14, ICEWS05-15, and GDELT show that TFOLQ outperforms state-of-the-art baselines in most of the temporal FOL queries.