Question answering over temporal knowledge graphs based on hierarchical semantic extraction

Temporal Knowledge Graph Question Answering (TKGQA) aims to address problems with temporal constraints using a Temporal Knowledge Graph (TKG). Existing models often overlook the structural information within problems, making it challenging to effectively manage complex temporal issues. To overcome these challenges, we propose the HSTQA model, which employs a hierarchical semantic extraction strategy to precisely capture semantic and implicit temporal information in queries. A multi-granularity fusion technique enhances the model’s capability to tackle problems with varying temporal granularities. Additionally, HSTQA incorporates graph convolutional networks to efficiently identify and integrate implicit temporal features and relational patterns. Extensive testing on the MultiTQ dataset shows that HSTQA improves Hits@1 by $\mathbf{1 0. 8 \%}$ compared to traditional methods, demonstrating its effectiveness in addressing multi-granularity temporal problems and enhancing complex temporal reasoning.