Research on Multi-hop Intelligent Question Answering Reasoning Methods Based on Knowledge Graphs

Abstract

In response to the challenge of capturing the structural and semantic information embedded in relation paths in complex multi-hop knowledge reasoning questions, this paper introduces a novel Knowledge Graph Re-ranking Relation Path Network model (KG-ReRPM). Leveraging representation learning techniques, this model unifies the vectorized representation of knowledge graphs and natural language questions, mapping them into the same vector space. The study extracts relation path sets relevant to the queries from the knowledge graph and delves into learning the semantic and structural features of these paths through Transformer networks. Within the vector space, we utilize a reasoning mechanism to identify relation paths that best align with the semantics of the question, thereby generating candidate answers. Experimental results demonstrate the effectiveness of the proposed method.