Ph d thesis-ahsan_slidesv3

http://www.fao.org/aims/

Aligning Controlled vocabularies for enabling
semantic matching in a distributed knowledge
management system

Ahsan Morshed
Doctoral Candidate
University of Trento
ahsan.morshed@fao.org
PhD Supervisor: Professor Fausto Giunchiglia
fausto@dit.unitn.it

Ahsan Morshed, FAO 1 / 54


Publications (1-3)
 A. Morshed. Controlled Vocabulary Matching in Distributed Systems, at
BNCOD 2009 Conference,UK.
 A. Morshed and M. Sini. Aligning Controlled vocabularies: Algorithm and
Architecture at Workshop on Advance Technologies for Digital Libraries
2009, AT4DL, Trento, Italy.
 M. Sini, J. Keizer, G. Johannsen, A. Morshed, S. Rajbhandari and M.
Amirhosseini.The AGROVOC Concept Server Workbench System:
Empowering management of agricultural vocabularies with semantics at
International Association of Agricultural Information Specialists (IAALD),
France, 2010.



Publications (4-6)
 A. Morshed, G. Johanssen, J. Keizer and M. Zeng,. Bridging End Users’
Terms and AGROVOC Concept Server Vocabularies. International
Conference on Dublin Core and Metadata Applications (DC-2010),
Pittsburgh, USA, 2010 (submitted).

 A. Morshed, M. Sini and J. Keizer. Aligning Controlled Vocabularies using a
facet based approach. (Technical Paper at FAO).

 A. Morshed and R. Singh. Evaluation and Ranking of Ontology Construction
Tools (Technical Paper).



Agenda
 Background: the role of controlled vocabulary in semantic matching
 The overall goal: Aligning Controlled Vocabularies in a distributed
system
 A facet based matching
 An Architecture for matching system
 A running prototype for matching system
 Evaluation Methodology and Results
 Limitations and Related Works
 Conclusions and Future work



Some matching techniques

 Element Matching techniques
ex: edit distance
 Corpus-based techniques
ex: token or extension of classes
 Structure-based tecniques
ex: graph matching
 Knowledge-based techniques
ex: external resources



Some matching systems
 Cupid
- element level and structure level matching
 RiMOM
- based on edit distance and Vector distance
 FALCON-AO
- based on Linguistic and structure matching
 CTXMatch, S-match
-based on knowledge based



Some matching projects
 HILT (High Level Thesaurus Project)
-JISC funded project, UK
-to facilitate the cross-searching of distributed information
services by subject in a multi-schema environment.
-used datasets (e.g.,DDC,LCSH, IPSV, AAT)

 CAT to AGROVOC
 Dr. Chan chung
 64,638 Chinese terms, 51,614 descriptors and 13,024 non-
descriptors
 13,105 exact matches,11,408 BT match, 173 NT match, and 17,47
other matches



Some matching project
 OAEI 2007 (Ontology Alignment Evaluation Initiative) -
Food Track
- AGROVOC-NALT thesauri

System Alignment Alignment Type

Falcon-AO 15,300 exactMatch

RiMOM 18,420 exactMatch

X-SOM 6,583 exactMatch

DSSim 14,962 exactMatch

[Willem , 2008]


Matching in Distributed System

 Edutella
 Edutella is an open source project that creates an infrastructure for sharing
metadata in RDF format
 It applies the peer-to-peer model using the JXTA protocol

 Swap
 aims at overcoming the lack of semantics in current Peer-to-Peer system



Semantic Matching in Lighweight
ontologies
 To use of lightweight ontologies for matching purpose, all entities need to
agree on the exact meaning of the concepts.

 Descriptive lightweight ontologies
-used for defining the meaning of terms as well the nature and structure of a
domain.

 Classification lightweight ontologies
-used for describing, classifying, and accessing collection of document.
[Fausto et al.,2007]



Controlled Vocabulary (CV)
 A vocabulary stores words, synonyms, word sense definitions (i.e.
glosses), relations between word senses and concepts; such a
vocabulary is generally referred to as the Controlled Vocabulary (CV)
if choice or selection of terms are done by domain specialists [ahsan et
al.,2009]



Controlled Vocabulary
 General controlled vocabulary:
Example: Thesaurus, WordNet, Classification, Directories, Lightweight
Ontologies

 Subject specific controlled vocabulary (SSCV)

 Library of Congress and Authors List
 Uniform List

 Series List



Applications for managing
controlled vocabularies

 Traditional Controlled Vocabulary tools
Ex: Old Agrovoc Thesaurus

 Modern Controlled Vocabulary
Ex: AGROVOC Concept Server



AGROVOC Concept Server

-store concepts
-Edit concepts
-visualize the
concepts

modern controlled vocabulary
Ref: http://nais.cpe.ku.ac.th/agrovoc/



Applications for exploiting
controlled vocabularies

 Background Knowledge

 Document annotation

 Information retrieval and extraction

 Audio and Video retrieval



Challenges of Matching

 Factors of heterogeneity problem

 Time
 Place
 Structure
 Culture diversity
 Different vocabulary specialists



Challenges of Matching

 Different heterogeneity

 Syntactic heterogeneity
 Lexical heterogeneity
 Semantic heterogeneity
 Pragmatic heterogeneity
 Metadata heterogeneity

[Pavel, 2006 ]



Problem of CV



FACET
 A facet is like a diamond that consists of different faces.

 Its distinct features allow thesauri, classifications or taxonomies to
be organized in different ways.

 composed of collectively exhaustive aspects of properties or
characteristics of a domain.
For example, a collection of rice might be classified using cultural
and seasonal facets.
[Fausto et al.,2009] [ahsan et al.,2009]



Faceted Controlled vocabulary



Faceted Controlled vocabulary

Seasonal rice type Cultural rice type



Creation of a Facet
 Domain Analysis
 analysis of terms are done by consulting domain experts
 simple concept are identified.

 Term collections and organization
 terms are order according to their characteristic and meaningful sequence
ex: cow and milk form a facet called Diary system(part of relationship)
[Fausto et al., 2009]



Exisiting Metholodies

 PMEST : Personality(P), Matter(M), Energy(E), Space (S), and Time(T)
[Ranganathan]

 DEPA : Discipline(D), Entity (E), Property (P), Action(A)
[Bhattachary and Fausto et al., 2009 ]



Properties of facets

 Hospitalities

 Compactness

 Flexibility

 Reusability

 The Methodology

 Homogeneity

[Bhattachary and Fausto et al., 2009 ]



Concept Facet Matcher

 Based on DEPA model

 CF={mg,lg,R} Where, mg is more general concepts ,lg is less general
concepts, R is related concepts.

 Based on Element Lebel Matchers

[Ahsan, 2009 and Ahsan et al., 2009 ]




Algorithm 1 buildCFacet(CV)
for i = 0 to CV do
store cF (Mg,Lg;R)
end for
return cF

[Ahsan et al., 2009 ]




Algorithm 2 MatchingFacet(CV1,CV2)
cF1=BuildCFacet(CV1)
cF2=BuildCFacet(CV2)
for i = 0 to cF 1:size do
for j = 0 to cF 2:size do
cfmatcher=elementLevelMatcher(cF 1;cF2)
end for
end for

[Ahsan et al., 2009 ]



System Architecture



Data Model

Agrovoc database
Ref: http://aims.fao.org/website/Download/sub


DATA Model

CABI database

Ref: http://cabi.org



A Running Prototype
Search Sring

Validators/ domain
specialist s


An architecture for a semantic search



Running Prototype for search

user’s choice



Evaluation and Results

 A domain Expert
 Exact Match
 Partial Match
 No Match



Datasets
Comparision
Characteristics AGROVOC CAB
Tree leaves 29172 47805
Term counts 18200 32884
Single words 6842 11720
MultiWords 11358 21161
Hierarchy depth 7 14
multiple BT 2546 1207
redundant BT 57 76



Datasets

AGROVOC version 2007-08-10 2007-08-10

CABI version 2009-11-01 2009-11-01
AGROVOC term-leaves 35036 35036
CABI term-leaves 29172 29172
Coversion hierarchy hierarchy
Knowledge base WordNet 2.1 SWN 400.000



Datasets

Relationship
BT NT RT UF

AGROVOC 228466 228424 326389 54370

CABI 15154 15841 41239 7094



Input files

Agrovoc input file CAB input file



Results
Facet based appraoch
Experiment 1 Experiment 2

Exact Match 5976 6021

Partial Match 164255 164278

No Match 69800745 69800745



Results
Standard Tool
Experiment 1 Experiment 2

Exact Match 8795 8795

Partial Match 334255 334258

No Match N/A N/A



Results
Min Max Min Max

Overall 25.8065 31.4496 21.7391 21.7391

Positive 18.6047 14.0814 10.4895 14.6154

Negative 97.1831 52.1495 94.7368 99.1304



Advantage of Facet based
System

 No knowledge base required

 Based on hidden semantic. Semantic meaning retrived during the
processing



Limitations
 Structure Problems

 AGROVOC SQL Format and CABI Text Format
 Provided CABI file does not contain chemical and scientific concepts

 Term Variants
 In AGROVOC, we found ``frog farms" which should have been ``frog farming"
because ``frog farms" is used for ``frog culture" and BT is ``aquaculture". Also, we
found the abbreviated term ``UHT milk" (one kind of milk product) which should
have been "UHT milk".
 There were some ambiguous term which had different meanings, for example
``cutting" ( i.e., slicing of bread or meat) or ``cuttings" (i.e.,propagation material).
 there were some terms spells whose meaning is to difficult to capture, for
example “2.4.4-T”, “2.4.5-TP 2.4-D”, “2.4 DES”, “2.4 dinitrohenol”. Similarly, CABI
contained the term “4-H Clubs”. These terms did make sense during any
mapping experiments.



Limitations

 Domain expert
 To evaluate our results, we were able to find one domain expert from
FAO but we did not get any domain expert from CABI. The results may
have been different if we had another domain expert.
 Lack of consistency
 Since the relationships in thesauri lack precise semantics, they are
applied inconsistently, both creating ambiguity in the interpretation
of the relationships and resulting in an overall internal structure that
is irregulated and unpredictable



Limitations

 Limited automated processing
 Traditional thesauri are designed for indexing and query formulation by
people and not for automated processing. The ambiguous semantics that
characterizes many thesauri makes them unsuitable for automated
processing.



Related Works
 [Fausto et. al, 2004] apply element level matching techniques
for semantic matching
 [Stamou et.al] apply string matching techniques for ontology
matching
 [Karin Koogan Breitman et.al 2005] apply string matching
techniques for lighweight ontology matching
 [Paul Buitelaar et. al, 2009] apply string matching for linguistic
matching system
 [Maria Teresa Pazienza et.al, 2007] Apply string matching for
semi-automatic matching system



Conclusion and Future work
 To build the extended knowledge base



 Integrating Mapping into AGROVOC concept Server




 We have described the facet based matching system for a large dataset
 We have shown a running prototype for this system.

 The majority of this work was done under the supervision of the FAO and
the CABI. At the moment, a prototype is running at the FAO

 We will integrate this mapping file for searching purpose in AGROVOC
Concept Server.



Questions



References
 [Fausto et al., 2003]: F.Gunchiglia and P. Shvaiko. Semantic Matching
Ontologies and Distributed System workshop, IJCAL,2003
 [Fausto et al., 2004]: F. Gunchiglia, P. Shvaiko, and M. Yatskevich. S-
Match: An algorithm and an implementation of semantic matching.
In Proceedings of ESWS’04, 2004.
 [Fausto et al., 2004]: F.Gunchiglia and M. Yatskevich. Element level
semantic matching. In meaning Coordination and Negotiation
workshop, ISWC,2004
 [Pavel et al., 2006]: P. Shvaiko, F.Gunchiglia and M. Yatskevich.
Discovering missing background knowledge in ontology matching. In
17th European Conference on Artificial Intelligence (ECAI 2006),
volume 141,pages 382-386,2006



References (cont)

 [Fausto et al., 2007]: F.Gunchiglia and I. Zaihrayeu. Light weight
Ontologies . Technical report at DIT, University of Trento Italy, October
2007
 [Pavel et al., 2007]: P. Shvaiko, and J.Euzenate. Ontology matching.
Springer, 1st edition , 2007.
 [Fausto et al., 2004]: F.Gunchiglia and M. Yatskevich. Element level
semantic matching. In meaning Coordination and Negotiation
workshop, ISWC,2004
 [S.R. Ranganathan]: S.R. Ranganathan. Element of library classification.
Asia Publishing house



References (cont)
 [Fausto et al., 2009]: F.Gunchiglia, B. Dutta, and V. Maltese. Faceted
lightweight ontologies. In LNCS, 2009
 [Bhattachary 1979]: G. Bhattachary. POPSI: its foundamentals and
precedure based on a general theory of subject indexing language. In
Library Science with a slant to Documentation, volume 16, pages.
 [Pavel]: P. Shvaiko . Iterative schema-based semantic matching (PhD
thesis), Technical report DIT-06-10Pavel]: 2,December 2006.

 [morshed 2009]: A. Morshed and M. Sini. Aligning Controlled
vocabularies: Algorithm and Architecture at Workshop on Advance
Technologies for Digital Libraries 2009, AT4DL, Trento, Italy
 [Morshed 2009]: A. Morshed, M. Sini and J. Keizer. Aligning
Controlled Vocabularies using a facet based approach. (Technical Paper
at FAO).



Thank You


Ph d thesis-ahsan_slidesv3

Recommended

Recommended

More Related Content

Similar to Ph d thesis-ahsan_slidesv3

Similar to Ph d thesis-ahsan_slidesv3 (20)

Ph d thesis-ahsan_slidesv3