Towards benefit-based RDF source selection for SPARQL queries

The Linked Data cloud consists of a great variety of data provided by an increasing number of sources. Selecting relevant sources is therefore a core ingredient of efficient query processing. So far, this is either done with additional indexes or by iteratively performing lookups for relevant URIs. None of the existing methods takes additional aspects into account such as the degree of overlap between the sources, resulting in unnecessary requests. In this paper, we propose a sketch-based query routing strategy that takes source overlap into account. The proposed strategy uses sketches and can be tuned towards either retrieving as many results as possible for a given budget or minimizing the number of requests necessary to retrieve all or a certain fraction of the results. Our experiments show significant improvements over state-of-the-art but overlap-ignorant methods for source selection.