Leveraging domain context for question answering over knowledge graph

This paper focuses on the problem of question answering over knowledge graph (KG-QA). With the increasing availability of different knowledge graphs in a variety of domains, KG-QA becomes a prevalent information interaction approach. Current KG-QA methods usually resort to semantic parsing, retrieval or neural matching based models. However, current methods generally ignore the rich domain context, i.e., category and surrounding descriptions within the knowledge graphs. Experiments shows that they can not well tackle the complex questions and information needs. In this work, we propose a new KG-QA approach, leveraging the domain context. The new method designs a neural cross-attention QA framework. We incorporate the new approach with question and answer domain contexts. Specifically, for questions, we enrich them with users' access log, and for the answers, we equip them with meta-paths within the target knowledge graph. Experimental study on real datasets verifies its improvement. The new approach is especially beneficial for domain knowledge graphs.