YARR! Yet Another Rewriting Reasoner Jrg Schnfisch ORE Workshop - PowerPoint PPT Presentation

YARR! Yet Another Rewriting Reasoner Jörg Schönfisch ORE Workshop 2013 Ulm, Germany, 22.07.2013

Agenda Introduction Query Rewriting Architecture & Implementation Performance Current development Introduction | Query Rewriting | Architecture | Performance | Current Development

Introduction Motivation for implementing YARR We needed a reasoner for closed source customer projects Requirements Stable implementation Liberal licensing Reuse of existing architecture Suitable for a collaborative and concurrent deployment scenario Introduction | Query Rewriting | Architecture | Performance | Current Development

Introduction YARR’s Features Query answering through query rewriting (Presto algorithm) Support of the OWL 2 QL profile Persistence in a relational database system (RDBMS) SPARQL endpoint Developed as part of a larger platform Introduction | Query Rewriting | Architecture | Performance | Current Development

Introduction YARR’s Limitations SPARQL 1.1 support limited to SELECT and ASK queries UPDATE, CONSTRUCT and DESCRIBE are not supported Property paths (especially negation of properties) not implemented Some built-in functions are not implemented, yet abs, ceil, floor, … Hashing functions Some String operations … Introduction | Query Rewriting | Architecture | Performance | Current Development

Introduction OWL 2 QL Profile Enables conjunctive queries to be answered in LogSpace using standard relational database technology Suitable for applications where relatively lightweight ontologies are used to organize large numbers of individuals and where it is useful or necessary to access the data directly via relational queries (i.e. SQL) Limitations: No cardinality restrictions No (disjoint) unions No transitive properties … Algorithms and prototypes for OWL 2 QL reasoning exist, but they are not easily available for commercial use in closed source products and customer projects Introduction | Query Rewriting | Architecture | Performance | Current Development

Query Rewriting Getting complete and sound answers from the database requires rewriting of the query to also fetch knowledge only implicitly defined by the TBox and the ABox Query Rewriting expands the query so that implicit facts are also retrieved from the ABox Features of Query Rewriting The ABox is not involved in the reasoning step during query answering Changes in the knowledge base are instantly reflected in query results Changes create no overhead (e.g. removal of old inferences) Only read access to the data is needed No preprocessing of the data required Trades less storage space for the knowledge base for more complex queries Introduction | Query Rewriting | Architecture | Performance | Current Development

Preliminaries Rewriting Example SPARQL Query: SELECT ?athlete ?games WHERE { ?event hasGoldWinner ?athlete ; occurredInGames ?games . ?athlete isRepresentativeOf Germany . } (Relevant) TBox axioms: InverseProperties(hasGoldWinner,wonGold) InverseProperties(isRepresentativeOf,hasRepresentative) Translation to Datalog: q(?games,?event) :- hasGoldWinner(?event,?athlete), occuredInGames(?event,?games), isRepresentativeOf(?athlete,Germany) Introduction | Query Rewriting | Architecture | Performance | Current Development

Preliminaries Rewriting Example Rewritten Datalog after applying the Tbox axioms: q(?games,?event) :- occuredInGames(?event,?games), view_1(?event,?athlete), view_2(?athlete) view_1(?event,?athlete) :- hasGoldWinner(?event,?athlete) view_1(?event,?athlete) :- wonGold(?athlete,?event) view_2(?athlete) :- isRepresentativeOf(?athlete,Germany) view_2(?athlete) :- hasRepresentative(Germany,?athlete) The translation to SQL results in a query with 24 subselects and altogether 50 join operations. Introduction | Query Rewriting | Architecture | Performance | Current Development

Preliminaries Query Rewriting vs Materialization Alternative approach to rewriting: Materialization Everything that can be inferred about the data is stored explicitly beforehand Queries can be executed as-is Changes in the data are more complex as they influence the inferences Advantages of Query Rewriting Changes to instances have no impact on the internal state of the reasoner Highly concurrent editing and reasoning Requires potentially less storage space Size of the ABox is irrelevant for the reasoning step Requires no additional steps during load time of the data Disadvantages Potentially more complex queries Introduction | Query Rewriting | Architecture | Performance | Current Development

Architecture Reasoner Architecture SPARQL Endpoint SPARQL to Datalog Query Rewriting Editing Services Datalog to SQL Persistence Mapper (any) RDBMS Introduction | Query Rewriting | Architecture | Performance | Current Development

Architecture Database Schema Based on the structure of the OWL 2 Metamodel Every part of the metamodel is stored in a separate table Facilitates editing of the knowledgebase Consists of 43 tables 6 tables for each type of entity 4 tables for each type of assertion 2 tables for literals 21 tables for different axioms (domain, range, equivalency, …) 10 auxiliary tables Introduction | Query Rewriting | Architecture | Performance | Current Development

Performance Sp2Bench SPARQL Benchmark 10k YARR 10k Stardog 250k YARR 250k Stardog 5M YARR 5M Stardog 10000,0000 1000,0000 100,0000 10,0000 sec 1,0000 0,1000 0,0100 0,0010 1 2 3a 3b 3c 4 5a 5b 6 7 8 9 10 11 12a 12b 12c Introduction | Query Rewriting | Architecture | Performance | Current Development

Current development Further support for SPARQL 1.1 Queries Negation in Property Paths Implement more built-in functions Support for CONSTRUCT Performance improvement Caching Optimizations in the SQL translator Analysis of slow queries Implementing adapters for ontology-based data access (OBDA) Possibly supporting R2RML (W3C mapping language from RDF to relational data) Introduction | Query Rewriting | Architecture | Performance | Current Development

Thank You!

Performance Sp2Bench SPARQL Benchmark 10k LSP 10k OWLIM 250k LSP 250k OWLIM 5M LSP 5M OWLIM 10000,0000 1000,0000 100,0000 10,0000 sec 1,0000 0,1000 0,0100 0,0010 0,0001 1 2 3a 3b 3c 4 5a 5b 6 7 8 9 10 11 12a 12b 12c

Architecture Pure Java Application UI built with Vaadin Export (OWL API) OWL Import & Faceted Search Visualizations A-Box Editing T-Box Editing Search UI Ontology … … … SPARQL Querying Editing Services Fulltext Search Services Services Lucene based OWL 2 QL Search Engine Query Rewriting CRUD Operations on RDBMS (Presto Algorithm), Persistence Mapper Sesame SPARQL Parser (Hibernate / jOOQ) Tested with Oracle, (any) Postgresql, RDBMS HSQLDB