CypherQA: Question-answering method based on Attribute Knowledge Graph

Abstract

In knowledge-based question answering(KBQA), most research adopts the question template matching, which faces with challenges such as unclear entity boundaries and difficult path inference when solving complex questions. In this paper, we propose a KBQA solution based on attribute graph. It extracts the mentions in text to recognize relations and entities, and transforms it into a slot-filling Cypher statement to query the answer. Meanwhile, we design a two-layer network based on a structural attention mechanism to optimize entity boundary identification. The solution provides new ideas of relation recognition for answering complex questions over attribute knowledge graph. Experimental results show that the proposed approach achieves promising performance on both CCKS2019 public dataset and the self-built vertical domain dataset.