A Learning-Based Semantic Approximate Query over RDF Knowledge Graph

Abstract

Large scale RDF knowledge graphs (KG) usually contain billions of labeled entities, and how to obtain the desired search results efficiently over such RDF KG for given SPARQL queries have attracted increasing attentions recently. However, it is difficult for users to write a complex SPARQL query without full knowledge of the underlying KG schema due to the "schema-free" feature of RDF KG (e.g. there are several ways to represent the similar knowledge). Furthermore, it is also a challenge to get all possible search results for a simple SQPRQL expression. In this paper, we study the problem of relaxing SPARQL queries over RDF KG in order to acquire semantic approximate results for an "incomplete query". The basic idea behind this paper is to conduct a text-based learning for achieving the semantics of entities and relations and then take use of these semantic information to generate the relaxed SPARQL query avoid user intervention. We first propose a corpus translating algorithm to transform the discrete RDF triples into the trainable corpus with high quality. And then we use the CBOW model to train the semantic vector for each entities and relations. Thirdly, we relax the SPARQL query by taking use of the semantic vectors and obtain the semantic approximate results over RDF KG. Experiments on DBpedia datasets confirm the superiority of our solution.