Semantic Web
&
Cased Based Reasoning
AIST Meeting JPL, CA 2003
Mehmet S. Aktas
[email protected]
Outline


Semantic Web Overview
 Semantic Web
 Motivations
 Ontology Languages
 Semantic Web and Cased Based Reasoning
Cased Based Reasoning Overview
 Cased Based Reasoning
 CBR Process
 Conversational Cased Based Reasoning
AIST Meeting JPL, CA 2003
Semantic Web Overview
“The Semantic Web is a major research initiative of the World Wide
Web Consortium (W3C) to create a metadata-rich Web of resources
that can describe themselves not only by how they should be
displayed (HTML) or syntactically (XML), but also by the meaning of the
metadata.”
From W3C Semantic Web Activity Page

“The Semantic Web is an extension of the current web in which
information is given well-defined meaning, better enabling computers
and people to work in cooperation.”

Tim Berners-Lee, James Hendler, Ora Lassila,
The Semantic Web, Scientific American, May 2001
AIST Meeting JPL, CA 2003
Motivations
Difficulties to find, present, access, or maintain
available electronic information on the web

Need for a data representation to enable software
products (agents) to provide intelligent access to
heterogeneous and distributed information.

AIST Meeting JPL, CA 2003
The Semantic Stack and Ontology Languages
B
OWL Full
DAML,
OIL,
DAML+OIL
A
OWL DL
OWL Lite
RDF Schema
RDF
XML, XML Schema
The Semantic Language Layer for the Web
From “The Semantic Web” technical report by Pierce
A = Ontology languages based on XML syntax
B = Ontology languages built on top of RDF and RDF Schema
AIST Meeting JPL, CA 2003
Resource Description Framework (RDF) - I
Resource Description Framework (RDF) is a framework for
describing and interchanging metadata (data describing the web
resources).

RDF provides machine understandable semantics for metadata.
This leads,

better precision in resource discovery than full text search,

assisting applications as schemas evolve,

interoperability of metadata.

AIST Meeting JPL, CA 2003
Resource Description Framework (RDF)- II

RDF has following important concepts

Resource : The resources being described by RDF are
anything that can be named via a URI.

Property : A property is also a resource that has a name, for
instance Author or Title.

Statement : A statement consists of the combination of a
Resource, a Property, and an associated value.
Example: Alice is the creator of the resource http://www.cs.indiana.edu/~Alice.
AIST Meeting JPL, CA 2003
The Dublin Core Definition Standard

RDF is dependent on metadata conventions for definitions.

The Dublin Core is an example definition standard which
defines a simple metadata elements for describing Web
authoring.

It is named after 1995 Dublin (Ohio) Metadata Workshop.

Following list is the partial tag element list for Dublin Core
standard.







Creator: the primary author of the content
Date: date of creation or other important life cycle events
Title: the name of the resource
Subject: the resource topic
Description: an account of the content
Type: the genre of the content
Language: the human language of the content.
AIST Meeting JPL, CA 2003
Example
Alice is the creator of the resource http://www.cs.indiana.edu/~Alice.
Property
Resource
Property
Value
creator
=
http://purl.org/dc/elements/1.1/creator
Alice
http://www.cs.indiana.edu/~Alice
• Property “creator” refers to a specific definition. (in this example by Dublin Core
Definition Standard). So, there is a structured URI for this property. This URI makes this
property unique and globally known.
• By providing structured URI, we also specified the property value Alice as following.
“http://www.cs.indiana.edu/People/auto/b/Alice”
Inspired from “The Semantic Web” technical report by Pierce
AIST Meeting JPL, CA 2003
Example
Why bother to use
RDF instead of XML?
Alice is the creator of the resource http://www.cs.indiana.edu/~Alice.
<rdf:RDF xmlns:rdf=”http://www.w3c.org/1999/02/22-rdf-syntax-ns##”
xmlns:dc=”http://purl.org/dc/elements/1.1”
xmlns:cgl=”http://cgl.indiana.edu/people”>
<rdf:Description about=” http://www.cs.indiana.edu/~Alice”>
<dc:creator>
<cgl:staff> Alice </cgl:staff>
</dc:creator>
</rdf:RDF>
• Information in the graph can be modeled in diff. XML organizations. Human readers would
infer the same structure, however, general purpose applications would not.
•Given RDF model enables any general purpose application to infer the same structure.
Inspired from “The Semantic Web” technical report by Pierce
AIST Meeting JPL, CA 2003
RDF Schema (RDFS )


RDF Schema is an extension of Resource Description Framework.
RDF Schema provides a higher level of abstraction than RDF.






It resembles
objected-oriented
programming
specific classes of resources ,
specific properties,
and the relationships between these properties and other resources can be
described.
RDFS allows specific resources to be described as instances of more
general classes.
RDFS provides mechanisms where custom RDF vocabulary can be
developed.
Also, RDFS provides important semantic capabilities that are used by
enhanced semantic languages like DAML, OIL and OWL.
AIST Meeting JPL, CA 2003
Limitations of RDF/RDFS

No standard for expressing primitive data types such as integer, etc.
All data types in RDF/RDFS are treated as strings.

No standard for expressing relations of properties (unique, transitive,
inverse etc.)

No standard for expressing whether enumerations are closed.

No standard to express equivalence, disjointedness etc. among
properties
AIST Meeting JPL, CA 2003
DAML, OIL and DAML+OIL - I

RDF\RDFS define a framework, however they have limitations. There is a
need for new semantic web languages with following requirements
 They should be compatible with (XML, RDF/RDFS)
 They should have enough expressive power to fill in the gaps in
RDFS
 They should provide automated reasoning support

Ontology Inference Layer (OIL) and DARPA Agent Markup Language
(DAML) are two important efforts developed to fulfill these requirements.

Their combined efforts formed DAML+OIL declarative semantic language.
AIST Meeting JPL, CA 2003
DAML, OIL and DAML + OIL - II

DAML+OIL is built on top of RDFS.
 It uses RDFS syntax.
 It has richer ways to express primitive data types.

DAML+OIL allows other relationships (inverse and transitivity) to be
directly expressed.

DAML+OIL provides well defined semantics, This provides followings:
 Meaning of DAML+OIL statements can be formally specified.
 Machine understanding and automated reasoning can be supported.
 More expressive power can be provided.
AIST Meeting JPL, CA 2003
Example
How is DAML+OIL is
different than RDF/RDFS?
Example: T. Rex is not herbivore and not a currently living species.
 This statement can be expressed in DAML+OIL, but not in RDF/RDFS
since RDF/RDFS cannot express disjointedness.

DAML+OIL provides automated reasoning by providing such expressive
power.
For instance, a software agent can find out the “list of all the carnivores that
won’t be any threat today” by processing the DAML+OIL data representation
of the example above.
 RDF/RDFS does not express “is not” relationships and exclusions.

From “The Semantic Web” technical report by Pierce
AIST Meeting JPL, CA 2003
Web Ontology Language (OWL)





Web Ontology Language (OWL) is another effort developed by the OWL
working group of the W3Consorsium.
OWL is an extension of DAML+OIL.
OWL is divided following sub languages.
 OWL Lite
 OWL (Description Logics) DL
 OWL Full – limited cardinality
OWL Lite provides many of the facilities of DAML+OIL provides. In
addition to RDF/RDFS tags, it also allows us to express equivalence,
identity, difference, inverse, and transivity.
OWL Lite is a subset of OWL DL, which in turn is a subset of OWL Full.
AIST Meeting JPL, CA 2003
From Semantic Web to Cased Based
Reasoning

Developing new tools, applications and architectures on top of the
Semantic Web is the real challenge.

AI techniques should be used to utilize the Semantic Web up to its
potentials.

CBR is an AI technique based on reasoning on stored cases.

CBR technique can be applied to do intelligent retrieval on metadata
of codes related Earthquake Science.
AIST Meeting JPL, CA 2003
Overview of Case-Based Reasoning
What is CBR?

CBR is reasoning by remembering: It is a starting point for new
reasoning

Problem-solving: CBR solves new problems by retrieving and
adapting records from similar prior problems.

Interpretive/classification: CBR understands new situations by
comparing and contrasting them to similar situations in the past

Case-based reasoning is a methodology of reasoning from specific
experiences, which may be applied using various technologies
(Watson 98)
AIST Meeting JPL, CA 2003
What is CBR?
Everyday Examples of CBR

Remembering today’s route from the place you live to campus and
taking the same route.

Diagnosing a computer problem based on a similar prior problem.

Predicting an opponent’s actions based on how they acted under
similar past circumstances

Assessing a hiring candidate by comparing and contrasting to
existing employees
AIST Meeting JPL, CA 2003
CBR Process

What is a Case?
 Input cases are descriptions of a specific problem.
 Stored cases encapsulate previous specific
problem situations with solutions.
 Another way to look at it:
 Stored cases contain a lesson and a specific
context where the lesson applied.
 The context is used to determine when the
lesson may apply again.
AIST Meeting JPL, CA 2003
CBR Process
When and how are cases used?
Given a Problem Description (P.D.) to be solved,
CBR follows a cyclical process.
 REtrieve the most similar case(s)
 REuse the case(s) to attempt to solve the problem
 REvise the proposed solution if necessary
 REtain the new solution as a part of new case.

AIST Meeting JPL, CA 2003
CBR Process
The CBR Cycle
Problem
Retrieve
Case-Base
Reuse
Retain
Revise
Confirmed solution
Proposed solution
AIST Meeting JPL, CA 2003
Conversational CBR (CCBR)

CCBR is a method of CBR where user interacts
with the system to retrieve the right cases.

System responds with ranked cases and
questions at each step

Question-answer-ranking cycle continues until
success or failure
AIST Meeting JPL, CA 2003
Conversational CBR

CCBR facilities
 Question
management facility
 Case management facility
 GUI for user-system interaction
 Facilities to display questions or cases
AIST Meeting JPL, CA 2003
A Prototype CCBR Application
AIST Meeting JPL, CA 2003
A Prototype CCBR Application

Purpose




Intelligent retrieval on metadata describing codes written for
earthquake science.
Guidance on how to run the codes to get reasonable results.
Guidance for inexpert users to browse and select codes
Casebase



disloc - produces surface displacements based on multiple
arbitrary dipping dislocations in an elastic half-space
simplex - inverts surface geodetic displacements to produce fault
parameters
VC - simulates interactions between vertical strike slip faults.
AIST Meeting JPL, CA 2003
A Prototype CCBR Application

Classification
Initial effort – dummy cases created to classify the different codes
 A general approach is needed

AIST Meeting JPL, CA 2003
A Prototype CCBR Application
CCBR CASE
Problem
Feature
Feature
=
Feature
Solution
<Question, Answer>
AIST Meeting JPL, CA 2003
A Prototype CCBR Application
How does Case Ranking take place in CCBR?





Retrieved cases are sorted based on their consistency
with the query case.
As the questions are answered more cases are
eliminated.
A case is ruled out only if there is a conflict between the
case and the query case
Consistency number for a case remains same if the case
has no answer for the question.
Consistency number for a case gets incremented if the
case has the same answer to the question as the query
case.
AIST Meeting JPL, CA 2003
A Prototype CCBR Application
CCBR CASEBASE
A Case from
CASEBASE
A
Feature 1
Feature 2
Feature 3
Feature 4
Case
Query Case
Feature 1
Feature 2
Feature 5
IF ((A.Feature1.Solution = B.Feature1.Solution) &
(A.Feature2.Solution = B.Feature2.Solution))
THEN Consistency # = 2
Case
=
<Problem, Solution>
AIST Meeting JPL, CA 2003
B
A Prototype CCBR Application
How does question ranking take place in CCBR?




Questions can be ranked based on their frequency factor
Questions can be ranked based on predefined inference
rules
Only distinguishing questions are to be ranked
Questions can be YES/NO questions, multiple choice
questions or questions with numerical answers.
AIST Meeting JPL, CA 2003
References

W3C Semantic Web Activity Page. Available from
http://www.w3.org/2001/sw/.

T. Berners-Lee, J. Hendler, and O. Lassila, “The Semantic Web.”
Scientific American, May 2001.

Resource Description Framework (RDF)/W3C Semantic Web Activity
Web Site: http://www.w3.org/RDF/.

D. Brickley and R. V. Guha (eds), “RDF Vocabulary Description
Language 1.0: RDF Schema.” W3C Working Draft 23 January 2003.

The DARPA Agent Markup Language Web Site: http://www.daml.org.

OIL Project Web Site: http://www.ontoknowledge.org/oil
AIST Meeting JPL, CA 2003
References






CBR on the web
http://www.cbr-web.org
Case-Based Reasoning Resources
http://www.aaai.org/Resources/CB-Reasoning/cbr-resources.html
AI Topics - CBR
http://www.aaai.org/AITopics/html/casebased.html
A mailing list including announcements, questions, and discussion about
CBR, managed by Ian Watson [email protected]
Riesbeck & Schank, Inside Case-Based Reasoning, Erlbaum, 1989.
Kolodner, Case-Based Reasoning, Morgan Kaufmann, 1993.
AIST Meeting JPL, CA 2003
Descargar

Semantic Web Overview