RarKGQA: Multi-hop Question and Answering Method Based on Knowledge Graph Embedding

Abstract

Graphs (KGQA) aims to derive the correct answer entities through multi-hop reasoning on Knowledge Graphs (KG). In recent years, embedding-based KGQA has emerged as an effective method to address KG sparsity. The KG embedding model serves as the foundation for embedding-based KGQA methods, and its effectiveness directly impacts the accuracy of question answering. Existing studies typically input the topic entities directly into the scoring function without considering the relation edge information of the topic entities in KG. Moreover, the extraction of semantic information contained in questions is also a crucial aspect. To address these issues, we propose a new model, RarKGQA. The model enhances KG embedding and question feature extraction while paying attention to the relationship side information of the topic entity in KG. We conduct extensive experiments on two benchmark datasets, showing that our method significantly outperforms state-of-the-art methods in its category. Furthermore, comprehensive ablation experiments validate the effectiveness of our method for the multi-hop KGQA task.