Event-centric hierarchical hyperbolic graph for multi-hop question answering over knowledge graphs

Question Answering over Knowledge Graphs (KGQA) blends natural language processing with structured knowledge representation. While much attention of existing research has been given to entity-centric representations, the significance of events has not been fully explored. This paper introduces a novel Event-centric Hierarchical Hyperbolic Graph system for KGQA that effectively integrates entity and event information from knowledge graphs. Utilizing hyperbolic geometry, our model captures hierarchical structures, offering a refined representation of questions and related knowledge. Additionally, our integration of a Hierarchical Graph Attentive Network (HGAT) with Contrastive Representation Learning enables our model to effectively extract deep semantics and align them with knowledge graph structures. Empirical evaluations on the EventQA dataset demonstrate our approach's effectiveness, significantly surpassing current leading models by 3% F1 and accuracy. This work not only extends the scope of KGQA but also highlights the importance of event-centric representations in knowledge-based tasks.