REDD: An Algorithm for Redundancy Detection in RDF Models

Abstract

Abstract. The base of Semantic Web specifications is Resource Description Framework (RDF) as a standard for expressing metadata. RDF has a simple object model, allowing for easy design of knowledge bases. This implies that the size of knowledge bases can dramatically increase; therefore, it is necessary to take into account both scalability and space consumption when storing such bases. Some theoretical results related to blank node semantics can be exploited in order to design techniques that optimize, among others, space requirements in storing RDF descriptions. We present an algorithm, called REDD, that exploits these theoretical results and optimizes the space used by a RDF description. MotivationThe realization of the Semantic Web (SW) vision [1] needs ontologies for generating or interpreting (semantic) metadata for resources. It is fundamental to have ontology creation and integration steps in order to share structural knowledge between ontology designers and users. Ontologies are to be expressed in RDF according to SW specifications, using languages such as RDFS 1 and OWL. 2 It is important to note that both RDFS and OWL ontologies can be expressed as RDF graphs, so that ontologies can be treated exactly as other RDF models. In RDF design, the least power principle was applied: data structures are to be kept as simple as possible. This imposes to have very simple basic components, that are URIs 3 , blank nodes and statements (or triples). These design decisions have the drawback that RDF descriptions tend to grow fast as the complexity of the knowledge they represent increases. This observation encourages SW research to investigate toward the most effective storage solutions for RDF knowledge bases, in order to minimize required space. Intuitively, the lesser the number of triples a software (say, a query engine) has to examine, the faster it will process them.

On the Termination of the Chase Algorithm

“…Although the results that are produced are homomorphically equivalent to the original graph, they are not necessarily lean. Our work does not subsume Esposito et al, 2005] nor the other way round. We focus on dierent aspects of redundancy elimination.…”

“…For example, in the graph {(a 1 , a 2 , a 3 ), (X, a 2 , Y )} the triple (X, a 2 , Y ) can be eliminated (X, Y are blank nodes) because both X and Y are treated like existentially quantied variables in the RDF semantics [World Wide Web Consortium, 2003c] and the triple (a 1 , a 2 , a 3 ) witnesses the existence of such a resource (X, a 2 , Y ). In Esposito et al, 2005] dierent algorithms are introduced that approximately compute a lean version of a given RDF graph. The notion of a lean graph is orthogonal to derivability by application-specic rules.…”

“…As RDF has only very simple language constructs, RDF data often becomes large. There has been a line of research [World Wide Web Consortium, 2003c;Gutierrez et al, 2004Gutierrez et al, , 2003Esposito et al, 2005] to minimize RDF graphs without losing any information, i.e. retaining homomorphic equivalence.…”

“…In Esposito et al, 2005] dierent algorithms are introduced that approximately compute a lean version of a given RDF graph. The notion of lean is orthogonal to derivability by application-specic rules.…”

“…The work in Esposito et al, 2005] eliminates triples in the sense of lean graphs and homomorphic equivalence. Although the results that are produced are homomorphically equivalent to the original graph, they are not necessarily lean.…”