- From: Andrew Newman <andrewfnewman@gmail.com>
- Date: Thu, 9 Nov 2006 05:52:25 +1000
- To: public-rdf-dawg-comments@w3.org
I'd like present to the DAWG public comments list my Honours thesis. It discusses a formal model, the relational model, to SPARQL. It builds on the work of by Cyganiak, Frasincar et al., Harris and Shadbolt, Pérez et al. and others. Hopefully, it's appropriate to some of the current discussions. It's available at (~500K): http://jrdf.sourceforge.net/RelationalBasedSPARQL.pdf In it I suggest that the current SPARQL specification is directly influenced by implementation specifics such as SQL and not RDF. It is argued that the semantics of SQL is a poor match to the RDF data model. Examples of this mismatch include: * The existence of NULL (section 2.3). * UNION and other operations may or may not return duplicates (section 2.4). * Lack of Compositional Semantics (section 2.5). * Order dependent OPTIONAL (like SQL's left outer join) (section 2.6). Outcomes presented include: * A way of mapping RDF and SPARQL operations to the relational model (section 4). * Using tuple subsumption to implement UNION and OPTIONAL using previous optimisation techniques (section 2.7) that is up to twice as fast as an alternate implementation (using join, antjoin and union) and up to 8 times faster than ARQ (section 4.5). * An order independent version of OPTIONAL using full outer join and tuple subsumption (section 4.4). Suggested future work includes: * Using SQL to implement tuple subsumption OPTIONAL and UNION (section 5.3) * Alternative ways of implementing ASK and CONSTRUCT (section 5.1) using the relational model as a basis. * Aggregate functions (section 5.1). * Other optimisation techniques if compositional semantics are chosen (section 5.2). The current code is only available through SF subversion: svn co https://svn.sourceforge.net/svnroot/jrdf jrdf
Received on Wednesday, 8 November 2006 19:52:39 UTC