Instance Store [PPT] - CO-ODE

arghtalentData Management

Jan 31, 2013 (4 years and 7 months ago)

130 views

Copyright © 2005, The University of Manchester

1

OWL Instance Store

reasoning over large volumes of
individuals

Daniele Turi


S Bechhofer, I Horrocks, L Li, P Lord, U Sattler


Protégé Tutorial, 30
th

June 2005

Copyright © 2005, The University of Manchester

2


Semantic web and bioinformatics applications:



millions of individuals



Existing reasoners cannot cope





instance Store


Scalable Ontological Reasoning

Copyright © 2005, The University of Manchester

3


10,000s classes (gene terms)



1,000,000s individuals (gene products)



Example individual: 1433_CANAL


corresponding description:


takesPartIn

IntracellularSignallingCascade

and

isPartOf

Chloroplast

and

hasFunction

ProteinDomainSpecificBindingActivity



Example: Gene Ontology

Copyright © 2005, The University of Manchester

4



Classes and properties


more or less
static




Individuals:
dynamic


Frequent updates


Concurrency


Persistency


Security

Schema v Instance Data

in ontologies

Schema

Instance Data

Copyright © 2005, The University of Manchester

5


Main observation:


Reasoning over
individuals

can
often

be
reduced to reasoning over their
descriptions
(i.e. the classes they belong to)



class
-
level reasoner + relational database




Schema




Instance Data



Use reasoner only when strictly necessary



Cache inferred schema information in database

Instance Store

Copyright © 2005, The University of Manchester

6


Stores
assertions


assert(Individual I, Description D)



Answers
retrieval

questions


retrieve(Description Q): Set of Individuals

Instance Store

basic functionality

Copyright © 2005, The University of Manchester

7

(
individual, description
)

assert

retrieve
(
description
)

Database

(MySQL,

Oracle,

PostgreSQL,

Hypersonic, etc)


OWL
Ontology

individual 1

individual 2

individual 3



Class
-
level Reasoner

(FaCT, FaCT++, Racer, etc)

Java

schema

instance

data

+

inferred

schema

Copyright © 2005, The University of Manchester

8

Cache descriptions used in assertions and queries



Descriptions

table



Assertions

table


individual


description id



Types
,
Equivalents
,
Children
,
Parents

tables


concept


description id

Relational Database

Oracle

MySQL

Hypersonic

Copyright © 2005, The University of Manchester

9

Copyright © 2005, The University of Manchester

10


Relational databases




persistency

&
scalability



Java + relational databases




enterprise technology (J2EE)





concurrency

&
security





robust

Copyright © 2005, The University of Manchester

11


Reasoners do not speak OWL, but DIG



DIG is the de
-
facto standard XML language for
communicating with Description Logic reasoners



DIG Interface


RACER


FaCT


FaCT++




Class
-
level Reasoner

Copyright © 2005, The University of Manchester

12



OWL Wrapper


OWL to DIG (and back)


OWL Abstract Syntax

OWL

OWL API

DIG API

Copyright © 2005, The University of Manchester

13

Assertion (example)

jo
: Kid
and some

hasPet Cat

Types

1.
Animal

2.
CatOwner

3.
Kid

4.
Person

5.
PetOwner

6.
Young

Parents

1.
CatOwner

2.
Kid

Children
and

Equivalents
:



no new entry

Copyright © 2005, The University of Manchester

14

Copyright © 2005, The University of Manchester

15

Retrieval


Delegate most of work to database



Worst case: query
not

equivalent

to



either a
class in ontology



or the
conjunction of its parents



a subsumption test for each concept in the
following result set (parametric in parents
of query):

Copyright © 2005, The University of Manchester

16

SELECT DISTINCT

description
FROM

Descriptions


WHERE

Descriptions
.id
IN (


SELECT

descriptionId
FROM

Types



WHERE

concept =

p
arent
-
1


INTERSECT


...


INTERSECT


SELECT

descriptionId
FROM

Types



WHERE

concept =

parent
-
n


MINUS


SELECT

Types
.descriptionId
FROM

Types
,
Children


WHERE

Children
.concept =
Types
.concept


AND

Children
.descriptionId =

id
)

Copyright © 2005, The University of Manchester

17

Copyright © 2005, The University of Manchester

18

Work in Progress


No relation between individual:





Solution:
precompletion

reduces most inferences to inferences about
classes


Core precompletion component ready and
under testing

1433_CANAL

daniele_turi

usedBy

Copyright © 2005, The University of Manchester

19

Conclusions

Appropriate technologies for appropriate tasks


Reasoner only when strictly necessary


Database


scalability, performance (set operations)


Java + database


EJB


concurrency


security

Copyright © 2005, The University of Manchester

20

Related Work



Loading data into description reasoners

(Borgida and Brachman ’93)



Parka system

(University of Maryland ’95)




Semantic

indexing
” (Schmiedel ’94)



Copyright © 2005, The University of Manchester

21

System details

http://instancestore.man.ac.uk


Code available from SourceForge




Copyright © 2005, The University of Manchester

22

Demo


Enriched Gene Ontology (
GO
)


25,180
classes


4
properties



Instance data mined from GO SWISS
-
PROT
database (in June 03)


653,762
individuals



involving 48,584 distinct descriptions