HR-SKGs: Hyper-Relational Semantic Knowledge Graphs for Multi-hop Reading Comprehension

Multi-hop Reading Comprehension (RC) is a challenging task that requires models to integrate dispersed information and perform multi-step reasoning. In recent years, graph-based methods have shown promising performance on multi-hop RC tasks. However, they often overemphasize nodes and connection strengths, neglecting the rich semantic information carried by the node relations. To address these limitations, we propose a novel Hyper-Relational Semantic Knowledge Graphs (HR-SKGs) model, which constructs hyper-relational graphs by integrating node, relation, and topic information, capturing semantic features more precisely. Additionally, we introduce a topic-aware mechanism during paragraph retrieval to effectively filter paragraphs relevant to the question. In the graph reasoning phase, we propose a hyper-relational graph attention mechanism that integrates relation-aware and topic-aware approaches to optimize information flow and reasoning between nodes. Experiments demonstrate that our approach achieves significant performance improvements on the HotpotQA dataset, and outperforms other graph-based methods.