Generating SPARQL Queries for Data Discovery

The number of datasets available on the Web represents a huge space of meaningful information. When done manually, identifying the relevant portion of the data for a specific need is time-consuming. This paper presents an approach that identifies and extracts the relevant portion of RDF data sources for some application requirements described by a target schema. The identification process is based on a semi-supervised learning algorithm that iteratively computes for each class of the target schema the similarity between source entities and both the class description and the candidate instance patterns already identified for this class. The extraction process generates SPARQL queries for the identified candidate instance patterns. We present some evaluations showing the effectiveness of our approach.