Fuzzy Search of Knowledge Graph with Link Prediction

Abstract

In the world of knowledge graphs, only the written word is considered to be a fact. However, many knowledge graphs expressed in RDF are incomplete, and knowledge graph embedding and graph neural networks are used to compensate for the incompleteness. In this paper, we propose a query language, TranSPARQL, which is an extension of SPARQL to enable RDF fuzzy search using these neural network-based link predictions. One predictor can be described by a variable in the pattern of subject, property and object in TranSPARQL. And we describe two implementations, one using TransE for link prediction and the other a hybrid implementation using both TransE and graph neural networks.