MACRE: Multi-hop Question Answering via Contrastive Relation Embedding

Multi-hop question answering over knowledge graphs (KGs) is a crucial and challenging task as the question usually involves multiple relations in the KG. Thus, it requires elaborate multi-hop reasoning with multiple relations in the KG. Two existing categories of methods, namely semantic parsing-based (SP-based) methods and information retrieval-based (IR-based) methods, either suffer from complicated logic forms for covering diverse questions or fail to offer traceable reasoning. In this paper, we propose a novel approach for multi-hop question answering over KGs via contrastive relation embedding (MACRE), powered by contrastive relation embedding and context-aware relation ranking. An adaptive beam search is developed to deliver the answer by identifying the optimal weighted inferential chain, boosting the searching efficiency and alleviating error propagation. The proposed method offers both interpretable reasoning and powerful answering ability, unifying the strengths of SP-based methods and IR-based methods. Extensive experimental results on several benchmark datasets demonstrate the effectiveness of our method.