Query-Oriented Summarization of RDF Graphs

The Resource Description Framework (RDF) is a graphbased data model promoted by the W3C as the standard for Semantic Web applications. Its associated query language is SPARQL. RDF graphs are often large and varied, produced in a variety of contexts, e.g., scientific applications, social or online media, government data etc. They are heterogeneous, i.e., resources described in an RDF graph may have very different sets of properties. An RDF resource may have: no types, one or several types (which may or may not be related to each other). RDF Schema (RDFS) information may optionally be attached to an RDF graph, to enhance the description of its resources. Such statements also entail that in an RDF graph, some data is implicit. According to the W3C RDF and SPARQL specification, the semantics of an RDF graph comprises both its explicit and implicit data; in particular, SPARQL query answers must be computed reflecting both the explicit and implicit data. These features make RDF graphs complex, both structurally and conceptually. It is intrinsically hard to get familiar with a new RDF dataset, especially if an RDF schema is sparse or not available at all. In this work, we study the problem of RDF summarization, that is: given an input RDF graph G, find an RDF graph SG which summarizes G as accurately as possible, while being possibly orders of magnitude smaller than the original graph. Such a summary can be used in a variety of contexts: to help an RDF application designer get acquainted with a new dataset, as a first-level user interface, or as a support for query optimization as typically used in semistructured graph data management [4] etc. Our approach is query-oriented, i.e., a summary should enable static analysis and help formulating and optimizing queries; for instance, querying a summary of a graph should reflect whether the query has some answers against this graph, or finding a simpler way to formulate the query etc. Ours is the first semi-structured data summarization approach focused on partially explicit, partially implicit RDF graphs. In the sequel, Section 2 recalls RDF basics, and sets the