Expressive Query Answering For Semantic Wikis

Expressive Query Answering For Semantic Wikis

Jie Bao, Rensselaer Polytechnic Institute

baojie@cs.rpi.edu, http://www.cs.rpi.edu/~baojie

Outline Background: Semantic MediaWiki

General Design Issues: Semantics and Expressivity

Formalizing SMW with Datalog

Extending SMW Modeling and Query Languages

Implementations and Experimental ResultsJan 18, 20112

Semantic Wiki as a Data Store

Jan 18, 20113

Wiki DB Triple Store Online dataData Layer

Map Data EvaluationPublication

Management

Still many not yet

mentioned…

Wiki Layer

App.Layer

(Semantic) Wiki Scripting

Semantic Template

Semantic Query

(PHP, Javascript) Wiki Extensions

HaloExtension

Parser Function

Concept Modeling: RDF, Relational Modeling, Rules

Semantic Forms

ProjectManagement

Remote SemWiki

Group Info. Management

Semantic MediaWiki (SMW)

4

It is the most popular semantic wiki system extending MediaWiki

Mediawiki: What you edit what you see

Jan 18, 2011

Semantic MediaWiki

5

SMW: What you edit(Modeling Language)

what you see

typed link (property)

To author knowledge

Jan 18, 2011

Semantic MediaWiki

6

SMW: What you edit(Querying Language)

what you see

To retrieve knowledge

Jan 18, 2011

Why SMW? Low-cost solution for light-weight semantic

applications Integrated environment for modeling and

querying Simple to setup, easy to use

Can work with hundreds of other MW/SMW extensions Templating, Visualization, Editing, I/O,

Workflow… Access Control, Forms, Maps, SPARQL…

Jan 18, 20117

Expressivity (SMW 1.5.4) SMW-ML (Modeling Language)

category instantiation e.g., [[Category:C]] property instantiation e.g., [[P::v]] subclass, e.g., [[Category:C]] (on a category page) subproperty, e.g., [[Subpropety of:Property:P]] (on a property page)

SMW-QL (Query Language) conjunction: e.g., [[Category:C]][[P::v]] disjunction: e.g., [[Category:C]] or [[P::v]], [[A||B]] or [[P::v||w]] property chain: e.g., [[P.Q::v]] property wildcat: e.g., [[P::+]] subquery: e.g., [[P::<q>[[Category:C]]</q>]] inverse property e.g., [[-P::v]] value comparison, e.g. [[P::>3]][[P::<7]][[P::!5]]

Jan 18, 20118

However, we often need more expressivity

Modeling Domain and Range: “has author” is from “person”

to “document” Inverse property: “has author” <-> “author of” Transitive property: “part of” …

Query Negation: find cities that are not capitals Counting: find professors who advise more than 5

studentsJan 18, 20119

Extending SMW

Goal: offer additional expressivity without losing “wikiness” (i.e., collaborative, simple, easy to learn, informality-tolerate, and evolving-capable )

Jan 18, 201110

Design Issues: Semantics and Expressivity

Jan 18, 201111

Design Issue 1: Open or Close world?

OWL/DL -Like DB/Rule-Like

Jan 18, 201112

or

Design Issue 2: Expressivity Supported

A subset of OWL that Can be implemented using rules Is syntactically simple for common wiki users

Why not full OWL 2 RL or OWL 2 QL? Too complicated for most wiki users

Jan 18, 201113

Design Issue 3: Implementation Reuse existing tools if we can

Low learning curve: hide details from users; incremental changes from SMW

Portability: allow users to choose different backend stores (MySQL, SQL Server, etc.)

Fast enough for a typical semantic wiki (has < O(104) pages [1])

Jan 18, 201114

[1] http://semantic-mediawiki.org/wiki/Sites_using_Semantic_MediaWiki

Solution Formalizing SMW modeling and query

languages using datalog Descriptive, closed-world semantics Well-understood complexity and many known

optimizations

Implementation: leverage highly-optimized LP solvers for

reasoning, e.g., DLV, Clasp, and Smodels Reuse SMW UI for rendering query results

Jan 18, 201115

Expressivity

Modeling Language: a subset of OWL Prime (or RDFS++ named by others) rdfs:subClassOf, subPropertyOf, domain, range owl:TransitiveProperty, SymmetricProperty,

FunctionalProperty, InverseFunctionalProperty, inverseOf owl:sameAs, equivalentClass, equivalentProperty

Query Language: SMW-QL, plus Negation as failure Cardinality

Jan 18, 201116

Modeling SMW with datalog

Jan 18, 201117

Translation Rules for SMW-ML

Subproperty Subclass Class instance Property instance Redirection

P(x,y) :- Q(x,y) . C(x) :- D(x) . C(a) . P(a,b) . a=b.

Jan 18, 201118

Translation Rules for SMW-QL

{{#ask: [[Category:City]] [[capital of::+]] }}

result(x) :- City(x), capital_of(x, y) .

Jan 18, 201119

Translation Rules for SMW-QL

{{#ask: [[Category:A]][[p3::category:B]] or

[[p.p1.p2::<q>

[[Category:D]] or [[p1::<q>[[SomePage]]</q>]]

</q> ||!v ||<q>[[Category:E]]</q> ]]}}

result(x) :- _tmp0(x). _tmp0(x) :- A(x),

p3(x,x0), x0=category:B. _tmp0(x) :- p(x,x2),

p1(x2,x3), p2(x3,x1), _tmp9(x1).

_tmp9(x1) :- _tmp12(x1). _tmp12(x1) :- D(x1). _tmp12(x1) :- p1(x1,x4),

x4=SomePage. _tmp9(x1) :- thing(x), x !

=v. _tmp9(x1) :- E(x1).

Conjunction

Disjunction

Subquery

Inequality

Property chain

Jan 18, 201120

Extending SMW-ML and SMW-QL

Jan 18, 201121

SMW-ML+

[[Domain::C]] [[Range::C]] [[Type::Transitive]] [[Type::Symmetric]] [[Type::Functional]] [[Type::InverseFuncti

onal]] [[Inverse of::Q]]

C(x) :- P(x,y) C(y) :- P(x,y) P(x,y) :- P(x,z), P(z,y) P(x,y) :- P(y,x) SameAs(x,y) :-

P(z,x),P(z,y) SameAs(x,y) :-

P(x,z),P(y,z) Q(x,y) :- P(y,x)

Jan 18, 201122

On page “Property:P”

SMW-QL+ : Negations

{{#askplus: [[<>Category:C]] [[Category:D]]}}

{{#askplus: [[Category:C]] [[<>P::+]]}}

result(x) :- D(x), not C(x) .

result(x) :- C(x), #count{x: P(x,y)}<=0 .

Jan 18, 201123

SMW-QL+: (Non)qualified Cardinality

{{#askplus: [[>=3#P::+]]}}

{{#askplus: [[>=3#P::

<q>[[Category:D]]</q>]]

}}

result(x) :- thing(x),#count{x: P(x,y)}>=3 .

result(x) :- thing(x),#count{x: P(x,y),D(y)}>=3 .

Jan 18, 201124

For safeness

Theoretical Complexity

Jan 18, 201125

SMW RDF

SMW-ML NL-complete • NP-complete;• P-complete for grounded graph [Bruijn and

Heymans 2007]SMW-ML+ NL-complete

SMW-QL P-complete• In L without subqueries

(SPARQL) P-complete [Perez et al 2006]

SMW-QL+ P-complete

Recall that L NL P NP

Implementation and Experimental Results

Jan 18, 201126

Implementation Using DLV as the reasoner

Other LP solvers may be used as well

Two work modes File-based: reasoning based on a static dump

(snapshot) of wiki semantic data. Database-based:  reasoning based on a

shadow database via ODBC; Real-time changes of instance data will be updated.

Optimization Caching

Jan 18, 201127

Example:

Jan 18, 201128

Caching

Inverse property

Transitive property

Scalability: Data Complexity

Test machine: 2 * Xeon 5365 Quad 3.0GHz 1333MHz /16G / 2 * 1TB

Dataset: part of DBLP, 10,396 pages, 100,736 triplesJan 18, 201129

10k 20k 30k 40k 50k 60k 70k 80k 90k 100k0.0000.2000.4000.6000.8001.0001.2001.400

Query time(s)Query time(s)

Dataset size (triples)

Near linear

{{#askplus: [[Category:Person]] }}

Scalability: Query Complexity

1 2 3 4 5 6 7 8 9 100.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

Query time(s)Query time(s)

Subquery depth

Jan 18, 201130

Near constant

{{#askplus: [[Knows::<q>[[Knows::<q>[[Knows::<q>…</q>]]</q>]]</q>]] }}

Dataset: DBLP 100k triples

Scalability: Query Complexity

1 2 3 4 5 6 7 8 9 100.000

0.200

0.400

0.600

0.800

1.000

1.200

1.400

Query time(s)Query time(s)

Disjunctions

Jan 18, 201131

Near constant

{{#askplus: [[Knows::+]] or [[Knows::+]] or [[Knows::+]] …}}

Dataset: DBLP 100k triples

The SemanticQueryRDFS++ extension

Jan 18, 201132

http://www.mediawiki.org/wiki/Extension:SemanticQueryRDFS++

Conclusions and Future Work Formalizing SMW using datalog allows us to

analyze the reasoning complexity of SMW extend SMW modeling and query languages for an

expressive subset of OWL implement a SMW query engine based on DLV that

is scalable for typical uses.

Future Work Incremental reasoning Customized reasoning rules SPARQL <-> SMW-QL+ translations

Jan 18, 201133