Entailment Processing for Large RDF Data Sets Using GPU

In the Semantic Web, the Resource Description Framework (RDF) has become the standard representation to describe Internet resources. RDF data is structured in triples comprising a subject, a predicate and an object: the predicate defines the relation between subject and object. The RDF Schema (RDFS) extends raw RDF data with a standardized vocabulary to allow for entailment, e.g., type inheritance or type inference. Processing large RDF data sets, which are commonly stored as text files, is a time-intensive task: querying and entailment on RDF data requires a huge amount of computational power and storage. We propose TripleID – a framework for RDF querying and entailment processing. TripleID provides a novel, compressed file format for RDF data and utilizes Graphics Processing Units (GPUs) for accelerated, highly parallelized data processing. We demonstrate the advantages of our framework on real-world RDF data: TripleID reduces storage size for RDF data by up to 75% and accelerates querying and entailment processing up to 40 times as compared to the state-of-the-art tools that use conventional CPUs.