HyperMatch: Knowledge Hypergraph Question Answering Based on Sequence Matching

Yongzhe Jia, Jianguo Wei, Zirui Chen, Dawei Xu, Lifan Han, Yang Liu

Abstract: Question answering for automatic answer retrieval and knowledge hypergraphs for complex knowledge representations are currently two popular research areas. However, no knowledge hypergraph question answering method is available for answering complex questions. Secondly, current information retrieval-based methods cannot perform sequence ranking on a word basis when dealing with complex questions. They cannot add high weights to the relations of candidate entities related to the questions. We propose the HyperMatch method, which takes a single hyperedge as a unit and completes the extraction of candidate answer entities by sequence matching and multi-relation attention mechanism. Experiments show that HyperMatch can achieve a 9.44% improvement in the Hits@1 metrics. To the best of our knowledge, this is the first knowledge hypergraph question-answering method based on information retrieval.