Upload
mina
View
42
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Ontology Mapping in Pervasive Computing Environment. C.Y. Kong, C.L. Wang, F.C.M. Lau The University of Hong Kong. Outline. Introduction Literature review Proposed design Evaluation Conclusion and Future works. Pervasive Computing. - PowerPoint PPT Presentation
Citation preview
Ontology Mapping in Pervasive Computing Environment
C.Y. Kong, C.L. Wang, F.C.M. Lau
The University of Hong Kong
Outline
Introduction Literature review Proposed design Evaluation Conclusion and Future works
Pervasive Computing M. Satyanarayanan - An environment saturated with computing
and communication capability, yet so gracefully integrated with users that it becomes a “technology that disappears”.
Various information flows: User intent Resource discovery and query Context information
Different types of computers communicate Smart devices Surrogates Sensors Peer-to-peer communication
Infeasible to expect all computers to have the same semantics on different information. i.e. the vocabulary of the messages, which includes the name and valid values of message elements
XML A language commonly used for data exchange XML sets rules for syntax for structured documents but it
does not provide meanings for terms Same term may be used with different meaning in different
context Different term may be used for items that have the same
meaning Hence, human needs to be involved to agree on tag
names or mappings between roughly equivalent sets of tags in related standard=> Device interoperability ↓
A new language has been developed
Ontology Provide a formal, explicit specification of a shared
conceptualization of a domain that can be communicated between people and heterogeneous and widely spread application systems
A formal explicit description of concepts in a domain of discourse (classes), properties of each concept describing various features and attributes of the concept (slot) and restrictions on these properties
Provide meanings for terms when information exchange Bridge knowledge gaps between different domains Enable knowledge sharing in open and dynamic distributed
systems Allow devices and agents not expressly designed to work
together to interoperate (i.e. better device interoperability)
Ontology (cont) Example: Country ontology (Source ontology)
Example: Instance
Country
name
City
located_in
capital
Geographical Location
name
Land Boundary
neighbor_countrypart_of
has_boundary
Country
Japan
City
Tokyo
capital
Geographical Location
Asia
Land Boundary
Koreapart_of
has_boundary
Class/Concept
Properties
Relationship
Ontology Related Researches Context Broker Architecture (CoBrA) [University of Maryl
and, 2003] Defines a set of OWL ontologies called SOUPA (Standard Ontol
ogy for Ubiquitous and Pervasive Applications) Ontologies for agent, personal device, time, space, event, docu
ment and policy Enable communication between devices using defined ontologie
s GAIA [University of Illinois, 2002]
Defines a set of ontologies for active space such as entity and context information
Enable communication between devices using defined ontologies
Communications may involve terms from different ontologies
Hence, Ontology Mapping is introduced
Scenario
I want to find a resource/function
Proxy A
Request
--- --- ---
--- --- ---
--- --- ---
Concepts specified
by Ontology O1
Resource Description
--- --- ---
--- --- ---
Concepts specified by Ontology O2
Resource Description
--- --- ---
--- --- ---
Concepts specified by Ontology O3
Proxy B
Smart Space B
Smart Space A
Scenario
I want to find a resource/function
Request
--- --- ---
--- --- ---
--- --- ---
Concepts specified
by Ontology O1
Resource Description
--- --- ---
--- --- ---
Concepts specified by Ontology O2
Resource Description
--- --- ---
--- --- ---
Concepts specified by Ontology O3
Proxy B
Smart Space B
Ontology Mapping Given two ontologies O1 and O2, mapping one ontology o
nto another means that for each entity (concept, relation or instance) in ontology O1, we try to find a corresponding entity, which has the same intended meaning, in ontology O2
Ontology O1 Ontology O2
Literature Review Source-based
Mappings are done by comparing the similarity of the concepts based on the properties of the concepts and the structure of the ontology defined in the source ontologies
Example: PROMPT [Stanford, 2000] Work well for ontologies having a specialized terminology like m
edical ontology Matching accuracy decreases when mapping ontologies with mo
re general terminologies Instance-based
Mappings are done by comparing the similarity of the concepts based on the source ontologies and their instances
Example: FCA-Merge [University of Karlsruhe ,2001], GLUE [University of Illinois and University of Washington, 2002]
Structure and properties of the concepts are not taken into consideration
Accuracy varies based on the instance sets
New Challenges
Online mapping with partial ontology information
Efficiency Space limitation of devices Knowledge propagation
Proposed Design Partial Ontology Mapping
A device submits a resource or function request (an instance I1 of ontology O1)
A resource or function is present and described by O2
Map all the concepts used in I1 with the concepts in O2 Number of concepts to be mapped reduces More efficient
Ontology O1 Ontology O2
Instance
Proposed Design (cont)
Hybrid of source-based and instance-based ontology mapping Properties of the concept and its relationships with oth
er concepts are considered Instances are considered to increase accuracy Store evaluation results of instances to
avoid handling large number of instances at one time but, still provide a reasonable amount of instances for mappi
ng
Methodology Notation
O1: source ontology of the request instance O2: source ontology of the resource Ir: request instance
For each concept (Ci) appear in Ir, Find a set of candidate concepts in O2
For each candidate concepts (Cn) Compute the similarity of Ci and Cn
The candidate concept with maximum similarity degree is the mapped concept of Ci
History Records
Extraction of candidate concepts Compare the name similarity of Ci and every concept C’ i
n O2
For the k-highest name similarity concepts, denoted by C1..k
Ci namelen
bstringlongest sulen ' nameCi name, CSim
..kof C concepts ub-class)chidren (s..kof C concepts er class)parent (
ighbor of its ne with each..kcepts of Cmerged con..k with Clationshiphat has reconcepts t
..kC
et andidate sPossible c
1
1sup1
1
1
Similarity of concepts Ci and Cn Similarity is defined as
ninini
ni
ni
ni
CCPCCPCCP
CCP
CCP
CCP
,~~,,
,
(1)
21
,
2
,
1,UNUN
UNUNCCP
CnCiCnCi
ni
(2)
where
Ux: instance set of ontology Ox
UxCi,Cn: instance set of ontology Ox that contains concepts Ci and Cn
N(instance set): Number of instances in the instance set
How to find N(U1Ci,Cn), N(U
1Ci,~Cn) and N(U
1~Ci,Cn)?
(1 ) and (2) from “Learning to map between ontologies on Semantic Web”, 2002
Find N(U1
Ci,Cn) means finding the number of instances in U1
Ci that also contain Cn
Partition U1 into two sets. One set contains concept Ci (denoted U1
Ci) while the other set does not contain concept Ci (denoted U1
~Ci) Partition U2 into two sets. U2
Cn and U2~Cn
N(U1Ci,Cn) = N(U1
Ci)*StructSim(Ci,Cn)
where StructSim(Ci,Cn): structure similarity of Ci and Cn
N(U1Ci,~Cn) = N(U1
Ci) – N(U1Ci,Cn)
N(U1~Ci,Cn) = N(U1
Cn) – N(U1Ci,Cn)
Similarly, N(U2Ci,Cn), N(U2
Ci,~Cn) and N(U2~Ci,Cn)
N(U1Ci,Cn), N(U1
Ci,~Cn), N(U1~Ci,Cn)
Find N(U1
Ci,Cn) means finding the number of instances in U1
Ci that also contain Cn
Partition U1 into two sets. One set contains concept Ci (denoted U1
Ci) while the other set does not contain concept Ci (denoted U1
~Ci) Partition U2 into two sets. U2
Cn and U2~Cn
N(U1Ci,Cn) = N(U1
Ci)*StructSim(Ci,Cn)where StructSim(Ci,Cn): structure similarity of Ci and Cn
N(U1Ci,~Cn) = N(U1
Ci) – N(U1Ci,Cn)
N(U1~Ci,Cn) = N(U1
Cn) – N(U1Ci,Cn)
Similarly, N(U2Ci,Cn), N(U2
Ci,~Cn) and N(U2~Ci,Cn)
N(U1Ci,Cn), N(U1
Ci,~Cn), N(U1~Ci,Cn)
Find N(U1
Ci,Cn) means finding the number of instances in U1
Ci that also contain Cn
Partition U1 into two sets. One set contains concept Ci (denoted U1
Ci) while the other set does not contain concept Ci (denoted U1
~Ci) Partition U2 into two sets. U2
Cn and U2~Cn
N(U1Ci,Cn) = N(U1
Ci)*StructSim(Ci,Cn)where StructSim(Ci,Cn): structure similarity of Ci and Cn
N(U1Ci,~Cn) = N(U1
Ci) – N(U1Ci,Cn)
N(U1~Ci,Cn) = N(U1
Cn) – N(U1Ci,Cn)
Similarly, N(U2Ci,Cn), N(U2
Ci,~Cn) and N(U2~Ci,Cn)
N(U1Ci,Cn), N(U1
Ci,~Cn), N(U1~Ci,Cn)
Structure Similarity Compute the similarity between each pair of property of Ci (deno
ted by PCi) and property of Cn (dentoed by PCn)
Instance Similarity is the similarity of the content of the instances
Property Similarity
for x = 1 to number of properties of Cn
, StructSim(Ci,C
n)
tyceSimilari*Inswdatatype)Pdatatype, Sim(P* w
ycardinality,PcardinalitP*Simwnamename,PP*Simw
,PPSim
CnCi
CnCiCnCi
CnCi
tan43
21
21 tantan
2tan1tan
ce of Onsproperty iNce of Onsproperty iN
bstringlongest suN
ceof ins, content ce inscontent ofSim
)),((
Pr
CnCin PP * Simty x in Cof properifrequency average
ilarityoperty sim
Structure Similarity Compute the similarity between each pair of property of Ci (deno
ted by PCi) and property of Cn (dentoed by PCn)
Instance Similarity is the similarity of the content of the instances
Property Similarity
for x = 1 to number of properties of Cn
, StructSim(Ci,C
n)
tyceSimilari*Inswdatatype)Pdatatype, Sim(P* w
ycardinality,PcardinalitP*Simwnamename,PP*Simw
,PPSim
CnCi
CnCiCnCi
CnCi
tan43
21
21 tantan
2tan1tan
ce of Onsproperty iNce of Onsproperty iN
bstringlongest suN
ceof ins, content ce inscontent ofSim
)),((
Pr
CnCin PP * Simty x in Cof properifrequency average
ilarityoperty sim
Structure Similarity Compute the similarity between each pair of property of Ci (deno
ted by PCi) and property of Cn (dentoed by PCn)
Instance Similarity is the similarity of the content of the instances
Property Similarity
for x = 1 to number of properties of Cn
, StructSim(Ci,C
n)
tyceSimilari*Inswdatatype)Pdatatype, Sim(P* w
ycardinality,PcardinalitP*Simwnamename,PP*Simw
,PPSim
CnCi
CnCiCnCi
CnCi
tan43
21
21 tantan
2tan1tan
ce of Onsproperty iNce of Onsproperty iN
bstringlongest suN
ceof ins, content ce inscontent ofSim
)),((
Pr
CnCin PP * Simty x in Cof properifrequency average
ilarityoperty sim
Structure Similarity Compute the similarity between each pair of property of Ci (deno
ted by PCi) and property of Cn (dentoed by PCn)
Instance Similarity is the similarity of the content of the instances
Property Similarity
for x = 1 to number of properties of Cn
, StructSim(Ci,C
n)
tyceSimilari*Inswdatatype)Pdatatype, Sim(P* w
ycardinality,PcardinalitP*Simwnamename,PP*Simw
,PPSim
CnCi
CnCiCnCi
CnCi
tan43
21
21 tantan
2tan1tan
ce of Onsproperty iNce of Onsproperty iN
bstringlongest suN
ceof ins, content ce inscontent ofSim
)),((
Pr
CnCin PP * Simty x in Cof properifrequency average
ilarityoperty sim
Structure Similarity Compute the similarity between each pair of property of Ci (deno
ted by PCi) and property of Cn (dentoed by PCn)
Instance Similarity is the similarity of the content of the instances
Property Similarity
for x = 1 to number of properties of Cn
, StructSim(Ci,C
n)
tyceSimilari*Inswdatatype)Pdatatype, Sim(P* w
ycardinality,PcardinalitP*Simwnamename,PP*Simw
,PPSim
CnCi
CnCiCnCi
CnCi
tan43
21
21 tantan
2tan1tan
ce of Onsproperty iNce of Onsproperty iN
bstringlongest suN
ceof ins, content ce inscontent ofSim
)),((
Pr
CnCin PP * Simty x in Cof properifrequency average
ilarityoperty sim
Structure Similarity, StructSim(Ci,C
n) Compute the similarity between each pair of relationsh
ip of Ci (denoted by RCi) and relationship of Cn (dentoed by RCn)
Relationship Similarity
for x = 1 to number of relationships of Cn
Structure Similarity
typeRtypeR*Simw
ycardinalitRycardinalitR*SimwnameRnameR*Simw
RRSim
CnCi
CnCiCnCi
CnCi
,
),(,
,
3
21
)( ni and CCn concept onship x i of relatisimilarityaverage
rityhip simila*relationswsimilarity*property w 21
Structure Similarity, StructSim(Ci,C
n) Compute the similarity between each pair of relationsh
ip of Ci (denoted by RCi) and relationship of Cn (dentoed by RCn)
Relationship Similarity
for x = 1 to number of relationships of Cn
Structure Similarity
typeRtypeR*Simw
ycardinalitRycardinalitR*SimwnameRnameR*Simw
RRSim
CnCi
CnCiCnCi
CnCi
,
),(,
,
3
21
)( ni and CCn concept onship x i of relatisimilarityaverage
rityhip simila*relationswsimilarity*property w 21
Structure Similarity, StructSim(Ci,C
n) Compute the similarity between each pair of relationsh
ip of Ci (denoted by RCi) and relationship of Cn (dentoed by RCn)
Relationship Similarity
for x = 1 to number of relationships of Cn
Structure Similarity
typeRtypeR*Simw
ycardinalitRycardinalitR*SimwnameRnameR*Simw
RRSim
CnCi
CnCiCnCi
CnCi
,
),(,
,
3
21
)( ni and CCn concept onship x i of relatisimilarityaverage
rityhip simila*relationswsimilarity*property w 21
Structure Similarity, StructSim(Ci,C
n) Compute the similarity between each pair of relationsh
ip of Ci (denoted by RCi) and relationship of Cn (dentoed by RCn)
Relationship Similarity
for x = 1 to number of relationships of Cn
Structure Similarity
typeRtypeR*Simw
ycardinalitRycardinalitR*SimwnameRnameR*Simw
RRSim
CnCi
CnCiCnCi
CnCi
,
),(,
,
3
21
)( ni and CCn concept onship x i of relatisimilarityaverage
rityhip simila*relationswsimilarity*property w 21
Structure Similarity, StructSim(Ci,C
n) Compute the similarity between each pair of relationsh
ip of Ci (denoted by RCi) and relationship of Cn (dentoed by RCn)
Relationship Similarity
for x = 1 to number of relationships of Cn
Structure Similarity
typeRtypeR*Simw
ycardinalitRycardinalitR*SimwnameRnameR*Simw
RRSim
CnCi
CnCiCnCi
CnCi
,
),(,
,
3
21
)( ni and CCn concept onship x i of relatisimilarityaverage
rityhip simila*relationswsimilarity*property w 21
No. of instances
Estimate the similarity between ontology O1 and O2
where N(O1) and N(O2) are the number of concepts in O1 and O2
N(U1Cn)
, N(U1Cn)
21
21
,
ONON
conceptslar nameer of simitotal numbOOSim
2
221 ),( CU*NOOSim
History Records Caching mapping results
Increase efficiency Caching instance mapping results
Maintain a reasonable amount of instances for mapping
Increase accuracy and reduce space usage Popularity counters
Each property or relationship of a concept has a popularity counter
Act as a weight for the importance of the concept Increase accuracy and reduce space usage Knowledge accumulation
Knowledge propagation
Evaluation Programming language: Java 1.4.2 Ontology language: OWL (Ontology Web Language) Ontology Parser: Jena 2.1 Input source ontologies:
Semantic Web Research Community (SWRC) ontology: 24 concepts
Manually created a similar concept as SWRC ontology: 20 concepts
Request instance: 6 – 8 concepts Result
Proposed design
Source based
Accuracy 80% >90%
Efficiency 6s 10s
Proposed design
Instance based
Accuracy ~70% ~70%
Efficiency 6s 20s
Conclusion New challenges
Online mappingEfficiencySpace limitationKnowledge propagation
Partial ontology mapping Future work
ExperimentsContextResource instances selection
Q & A