Provenance-Based SPARQL Query Formulation

We present in this paper a novel solution for assisting users in formulating SPARQL queries. The high-level idea is that users write "semi-formal SPARQL queries" , namely, queries whose structure resembles SPARQL but are not necessarily grounded to the schema of the underlying Knowledge Base (KB) and require only basic familiarity with SPARQL. This means that the user-intended query over the KB may differ from the specified semi-formal query in its structure and query elements. We design a novel framework that systematically and gradually refines the query to obtain candidate formal queries that do match the KB. Crucially, we introduce a formal notion of provenance tracking this query refinement process, and use the tracked provenance to prompt the user for fine-grained feedback on parts of the candidate query, guiding our search. Experiments on a diverse query workload with respect to both DBpedia and YAGO show the usefulness of our approach.