Processing RDF Using Hadoop

TLDR: This work proposes a framework which is constructed using Hadoop to store and retrieve massive numbers of RDF triples by taking advantage of the cloud computing paradigm, and results confirm that the proposed framework offers multi-fold efficiencies and benefits which include on-demand processing, operational scalability, competence, cost efficiency and local access to enormous data.

Abstract: The basic inspiration of the Semantic Web is to broaden the existing human-readable web by encoding some of the semantics of resources in a machine-understandable form. There are various formats and technologies that help in making it possible. These technologies comprise of the Resource Description Framework (RDF), an assortment of data interchange formats like RDF/XML, N3, N-Triples, and representations such as RDF Schema (RDFS) and Web Ontology Language (OWL), all of which help in providing a proper description of concepts, terms and associations in a particular knowledge domain. Presently, there are some existing frameworks for semantic web technologies but they have limitations for large RDF graphs. Thus storing and efficiently querying a large number of RDF triples is a challenging and important problem. We propose a framework which is constructed using Hadoop to store and retrieve massive numbers of RDF triples by taking advantage of the cloud computing paradigm. Hadoop permits the development of reliable, scalable, proficient, cost-effective and distributed computing using very simple Java interfaces. Hadoop comprises of a distributed file system HDFS to stock up RDF data. Hadoop Map Reduce framework is used to answer the queries. MapReduce job divides the input data-set into independent units which are processed in parallel by the map tasks , which then serve as inputs to the reduce tasks. This framework takes care of task scheduling, supervising them and re-execution of the failed tasks. Uniqueness of our approach is its efficient, automatic allocation of data and work across machines and in turn exploiting the fundamental parallelism of the CPU cores. Results confirm that our proposed framework offers multi-fold efficiencies and benefits which include on-demand processing, operational scalability, competence, cost efficiency and local access to enormous data, contrasting the various traditional approaches.