Joint linking of entity and relation for question answering over knowledge graph

Entity linking and relation linking are two crucial components in many question answering systems over knowledge graphs, which aim to identify the relevant entity or relation mentions in a question and link them to the target entity or relation in the knowledge graph. Previous studies mostly solve these two tasks independently or as sequential tasks, which usually leads to poor performance since the short texts in most questions lack the context information needed for disambiguation. In this paper, we propose an approach to jointly perform entity linking and relation linking. The idea is to exploit both the independent and joint features of the candidates for disambiguation, which captures different characteristics when the knowledge graph information and the semantics of the question are both considered. We evaluated our approach on the QALD-7 and LC-QuAD datasets and the experimental results shows that our approach significantly outperforms the existing entity and relation linking approaches.