RDF Data Management

点击翻译按钮获取中文摘要

Graph data are of growing importance in many applications including the semantic web, social network analysis, bioinformatics, and physical communication networks. Graphs naturally model complicated structures in these fields, such as the relationships among people in a social network or the protein-protein interaction networks. The size and complexity of these graph data raise significant data management and data analysis challenges. My broad research scope is the study of these problems.**In this discovery grant, my focus is on the graph structures that arise from models of Web resources. The Resource Description Framework (RDF) is the standard (proposed by W3C) by which Web objects are commonly modeled. RDF is a self-descriptive data model that is suitable for machine understanding and interpretation, and, therefore, expected to facilitate the "semantic web". W3C has also defined a query language, called SPARQL, for accessing RDF repositories. RDF data sets have started to proliferate and grow. For example, Yago and DBPedia extract facts from Wikipedia and store them in RDF format to facilitate structural queries over Wikipedia; many local governments are now encoding the resources they provide to citizens in RDF format as part of the e-government initiatives; biologists have built elaborate RDF data collections (BioRDF and Uniprot RDF) for community sharing of experimental data; and Linked Open Data (LOD) initiative has been growing (as of September 2011 - which are the latest available information - over 31 million triples [tuples]) as a web data integration platform. Consequently, managing and analyzing large and distributed RDF datasets have emerged as an urgent and important concern.**My group's approach to RDF data management and analysis differs from many of the existing approaches that map, in one way or another, RDF into a relational representation and convert SPARQL queries into SQL. Although this has the advantage of leveraging mature technology, it gives rise to performance and modeling mismatch problems. We model an RDF dataset as a graph (which is the native model for RDF) and also represent a SPARQL query as a graph. Consequently, query execution reduces to graph matching. This approach has modeling and performance advantages. Within this general approach, I intend to study the following issues over the next five years:**1. Efficient storage structures for RDF graphs.*2. Efficient and effective query processing and optimization techniques for SPARQL queries (including aggregation queries that are now part of the SPARQL standard).*3. Distribution of RDF graphs and evaluation of SPARQL queries over distributed RDF stores.*4. Web data querying and integration using RDF, which requires some reasoning capability over RDF data (so called OWL 2 entailment regime).**The methodology that will be followed includes algorithmic studies, development of prototype systems, and extensive experimentation.**Successful completion of this research will result in the development of efficient and effective techniques for RDF data management, and web data integration and querying through RDF. Since RDF technology is now widely deployed (including by various levels of government at a number of countries), the results will have significant impact both technically and societally.