Abstract. The Semantic Web aims at building cross-domain and distributed databases across the Internet. SPARQL is a standard query language for such databases. Evaluating such queries is however NP-hard. We model SPARQL queries in a declarative way, by means of CSPs. A CP operational semantics is proposed. It can be used for a direct implementation in existing CP solvers. To handle large databases, we introduce a specialized and efficient light solver, Castor. Benchmarks show the feasibility and efficiency of the approach.

Abstract. Efficient evaluation of complex SPARQL queries is still an open research problem. State-of-the-art engines are based on relational database technologies. We approach the problem from the perspective of Constraint Programming (CP), a technology designed for solving NP-hard problems. Such technology allows us to exploit SPARQL filters early-on during the search instead of as a post-processing step. We propose Castor, a new SPARQL engine based on CP. Castor performs very competitively compared to state-of-the-art engines.