High-level Data Access Based on Query Rewritings

1
High-level Data Access Based on Query Rewritings
Ekaterina Stepalina
Higher School of Economics
2
High-Level Data Access

• Concentration on application domain tasks
• Abstraction from data sources
• Efficient work

• Research
• This problem is actively considered on modern
  scientific conferences on knowledge
  representation and ontologies OWLED (2009),
  (ICDE IIMAS, 2008) , the Semantic Web magazine
  (2011 the Mastro System)
• W3C developed OWL 2, OWL 2 QL (2008) and etc.

3
Ontology-Based Data Access (OBDA)

• Large amounts of data (distributed, inconsistent)
• Main task query answering (domain-oriented and
  efficient)

4
What is Ontology?

• Ontology is a knowledge domain described on some
  knowledge representation language.
• Entity-Relationship and UML Class diagrams can be
  seen as ontology languages.

5
Logic-Based Knowledge Representation

• Enables semantic processing of data
• Enables inference of implicit knowledge
• Well studied and actively developed
• Description logics (Baader,1999), esp. DL-Lite
• Standardized
• OWL 2 Profiles

6
DL-Lite Best Suites for OBDA

• High expressive and computationally efficient
• Allows delegating query answering to DBMSs and
  using all advantages of modern relational
  technologies
• Supported by the W3C standard - OWL 2 QL

7
Query Answering Problem

• Given a query and an n-tuple of objects
  from A. Decide, whether , or the
  n-tuple is the answer for with respect
  to K.
• For knowledge represented in DL-Lite, we can
  formulate queries in domain concepts, translate
  them into ordinary SQL queries and perform over
  separate databases.

8
OBDA System Architecture

• Ontology Editor
• OBDA-Enabled Reasoner
• Mapping Processor
• Data Source Manager
• Consistency Checker

9
Query Rewritings

• OBDA-Enabled Reasoner rewrites the initial
  ontology query into a set of UCQ (union
  conjunctive query).
• Mapping Processor builds an SQL from UCQ and
  given mappings.
• The initial query syntax may differ (SparQL,
  datalog query, etc.)

10
TBox and ABox in DL

• TBox is a finite set of concept and role
  inclusion axioms
• ABox is a finite set of assertions
• Where - objects name, A concept name, P
  role name, q integer.

11
Interpretation

• Interpretation (the particular instance of KB)
  is a pair if non-empty domain and an
  interpretation function
• , , and
  .
• UNA (unique name assumption)

12
OWL 2 QL

• UNA is ignored (in)equality must be defined
  explicitly
• Language expressive power reduced up to
• (other
  designation - ).
• Basic conceptual modeling relations are
  available (A)sym, (Ir)Ref, Tran
• Main constraints of
• Functional relations cannot be defined
• Particular roles cannot be assigned only to
  specific concepts, all roles are applied to all
  concepts
• Disjunction coverage of knowledge domain cannot
  be defined

13
Query Rewriting Sample

• RDB tables Person(name, age), Lives (person,
  city), Manages (boss, employee).
• Query Get the names and ages of all people
  living in the same city with their boss.
• UCQ
• Simplified UCQ
• SQL query
• SELECT P.name, P.age
• FROM Person P, Manages M, Lives L1, Lives L2
• WHERE P.nameL1.person AND P.nameM.employee AND
  M.bossL2.person AND L1.cityL2.city

14
Query Rewriting Algorithms

• CGLLR (Calvanese et al., 2007)
• - Applies all suitable TBox axioms to
• - Replaces axioms containing existential
  qualifications with another 3 axioms, which
  increases the number of UCQ
• RQR (Pérez-Urbina, Horrocks, Motik, 2009)
• Generates clauses from TBox assertions and then
  resolve clauses with query
• Potentially supports more expressive DLs

15
Query Rewriting Benchmark

• 9 ontologies with axioms,
  containing -existential qualification
• Vicodi (V)
• Stock exchange (S)
• University (U,UX)
• Adolena (A,AX)
• Synthetic (P1, P5,P5X)

16
Comparison Results

• RQR is more preferable to implement in
  OBDA-enabled reasoners, than CGLLR
• Generates less UCQ, especially for ontologies
  with large number of existential qualifications
• May be further optimized and advanced to more
  expressive DLs, than

17
Produced Rewritings
18
Running Time, ms
19
Current Work

• Preparing an ontology for a real application
  interactive television platform (IPTV) for
  testing algorithms on real data
• Optimizing RQR reducing the number of generated
  clauses
• Main idea not advance RQR, but support more
  expressiveness and all OWL 2 QL constructors in
  powerful mappings

20
References

• The Description Logic Handbook Theory,
  Implementation and Applications. Cambridge
  University Press, 2002. ISBN 0521781760. Edited
  by F. Baader, D. Calvanese, D. McGuinness, D.
  Nardi, P. F. Patel-Schneider.
• F. Baader. Logic-Based Knowledge Representation.
  In M.J. Wooldridge and M. Veloso, editors,
  Artificial Intelligence Today, Recent Trends and
  Developments, number 1600 in Lecture Notes in
  Computer Science, pages 1341. Springer Verlag,
  1999.
• Artale, A. Calvanese, D. Kontchakov, R. and
  Zakharyaschev, M. (2009) The DL-Lite family and
  relations. Journal of Artificial Intelligence
  Research 36 (1), pp. 1-69. ISSN 1076-9757.
• H.Perez-Urbina, I.Horrocks, and B.Motik. E?cient
  Query Answering for OWL 2. In Proceedings of the
  8th International Semantic Web Conference
  (ISWC2009), Chantilly, Virginia, USA, 2009.
• H.Perez-Urbina, B.Motik, and I.Horrocks.
  Tractable Query Answering and Rewriting under
  Description Logic Constraints. JournalofAppliedLog
  ic, 2009.

21
High-level Data Access Based on Query Rewritings

• Questions?