Ontology Matching Based on hypernym, hyponym, holonym, and meronym Sets in WordNet

International journal of Web & Semantic Technology (IJWesT), Vol.1, No.2, April 2010
1
ONTOLOGY MATCHING BASED ON HYPERNYM,
HYPONYM, HOLONYM, AND MERONYM SETS IN
WORDNET
JungAe Kwak and Hwan-Seung Yong
Department of Computer Science and Engineering,
Ewha Womans University, Seoul, Korea
jakwak@ewhain.net, hsyong@ewha.ac.kr
ABSTRACT
Considerable research in the field of ontology matching has been performed where information sharing
and reuse becomes necessary in ontology development. Measurement of lexical similarity in ontology
matching is performed using synset, defined in WordNet. In this paper, we defined a Super Word Set,
which is an aggregate set that includes hypernym, hyponym, holonym, and meronym sets in WordNet.
The Super Word Set Similarity is calculated by the rate of words of concept name and synset’s words
inclusion in the Super Word Set. In order to measure of Super Word Set Similarity, we first extracted
Matched Concepts(MC), Matched Properties(MP) and Property Unmatched Concepts(PUC) from the
result of ontology matching. We compared these against two ontology matching tools – COMA++ and
LOM. The Super Word Set Similarity shows an average improvement of 12% over COMA++ and 19%
over LOM.
KEYWORDS
Ontology Matching, Property Unmatched Concept, Semantic Relationship Set, Super Word Set Similarity
1. INTRODUCTION
An ontology consists of concepts, attributes, and properties representing relationships between
concepts. Ontology’s properties can represent user-defined relationships as well as is-a and has-
a relationships. Defining semantic relationships between concepts enables the development of a
machine system that can automatically interpret and understand the meaning of concepts used in
ontologies. For such automatic knowledge acquisition, ontologies have been widely studied and
developed in a variety of domains, which increases the need for sharing and reusing available
ontologies.
However, the task of sharing and reusing ontologies has many challenges. First, there is a
problem related to polysemy and synonym. A word can have multiple meanings (i.e., polysemy)
or there may be more than one word that means the same thing in a domain (i.e., synonyms),
which causes difficulty in ontology matching. Second, polymorphism in expressions also brings
about a difficulty in finding semantically related concepts. For example, an ontology might use
noun forms of names and another ontology uses adjectival forms of names to indicate the same
concept. Third, ontologies do not give a restriction on the count of a concept to represent a
meaning. For example, in order to represent someone’s name, an ontology simply has a
concept–‘Name’, while another ontologies use two concepts ‘FirstName’ and ‘LastName’ [1].
Finally, both a concept and a property can be used to give a meaning. For example, an ontology
can use the concept ‘adviser’ and another can use the property ‘advisedBy’ to indicate someone
who gives advice [1]. Due to such difficulties, ontology matching has been the subject of

2
considerable research. In particular, as users increasingly demand a way to automatically
exchange information and knowledge in applications, the automation of ontology matching
becomes a research issue of great importance [2].
An Ontology matching is generally comprised of three phases of similarity measure – similarity
measure between concepts, similarity measure between properties, and logical inference
similarity measure; and there are four major dimensions of similarity, i.e., lexical similarity,
structural similarity, instance similarity, and logical inference similarity. Existing ontology
matching methods rely heavily on lexical similarity. Lexical similarity generally employs the
character comparison method that focuses only on order of characters, irrespective of their
meanings. However, this approach cannot solve the problem of polysemy and synonym
described above. To address this drawback, lexical similarity in ontology matching is often
measured using Synonym Set(synset) defined in WordNet. If names of concepts or properties of
two ontologies are included in synset of WordNet, a pair of concepts or properties is
determined, which results in ontology matching in measuring the lexical similarity.
In this paper, we propose an ontology matching method using the Super Word Set Similarity.
The Super Word Set is an aggregate set that includes semantic relationships of words in
WordNet (i.e. hypernyms, hyponyms, holonyms, and meronyms).
We performed a new form of concept matching based on the rate of including names of
concepts and synset’s words in the Super Word Set. The proposed method first extracts the
Matched Concepts(MC) and the Matched Properties(MP) that are produced in the phase of
similarity measure between concepts and in the phase of similarity measure between properties.
Next, we extracted the Property Unmatched Concepts(PUC) that are not matched concepts
among the domain concepts and range concepts of the MPs. The Super Word Set Similarity
measure is then performed only between the PUCs, which eventually results in finding
additional MCs. In this way, the proposed method can achieve a higher matching rate than the
previous ontology matching methods.
This paper is organized as follows. Section 2 describes ontology matching tools and algorithms,
similarity measures used in ontology matching, and existing research works using synset in
WordNet. Section 3 presents the phase of ontology matching and the definitions of the terms
used in this paper. Section 4 states the proposed ontology matching method using the Super
Word Set Similarity measure, and the experiment results of the proposed method are given in
Section 5. Finally, Section 6 presents conclusions and future areas of work.
2. RELATED WORK
2.1. Ontology matching tools and algorithms
LSD (Learning Source Description) [1] employs a multi-strategy learning approach. In the
training phase, LSD chooses one learner among several base learners, such as Name Learner,
Content Learner, and XML Learner, based on their confidence score. In the matching phase,
ontology matching is performed using the learner chosen in the training phase, in order to
improve matching accuracy.
GLUE [3], a system extended from LSD, performs semi-automatic ontology matching. GLUE
creates a joint probability distribution table with neighbourhood-nodes, using the multi-strategy
learning approach. In addition, GLUE supports ontology matching using Relaxation Labeling.
MOMIS (Mediator Environment for Multiple Information Source) [4] creates a lexical matrix
using WordNet. A lexical matrix consists of rows storing word forms and columns storing word

3
meanings. Relations between words such as SYN (synonyms), BT (broader terms), NT
(narrower terms), and RT (related terms) can be represented using the developed lexical matrix.
MAFRA(An Ontology Mapping Framework for the Semantic Web) [5] provides a distributed
mapping process that consists of five horizontal and four vertical modules. It also performs
ontology matching using concept and property bridges.
LOM (Lexicon-based Ontology Mapping) [6] proposes four phases of a lexical similarity
measurement in ontology matching. We will describe these four phases in detail in Section 3.1.
COMA++ (Combining match algorithms) [7] is a well known ontology matching tool providing
a graphical user interface. COMA++ adopts multi-match strategies to perform matching of
relational schemas, W3C XSD, and OWL. The implemented multi-match strategies utilize
fragment-based matching and reuse-oriented matching.
Unlike the previous ontology matching methods focusing on 1:1 mapping, RiMOM (Risk
Minimization based Ontology Mapping) [8] automates the process of discoveries on 1:1, n:1,
1:null and null:1 mappings. Besides, RiMOM solves the problem of name conflicts in mapping
process using thesaurus and statistical techniques.
ILIADS (Integrated Learning In Alignment of Data and Schema) [9] proposes an ontology
matching algorithm based on logical inference. ILIADS integrates the previous ontology
matching approach and the logical inference similarity measure to achieve better matching of
ontologies.
2.2. Similarity measures in Ontology matching
In order to obtain ontology matching results, after each similarity in ontology matching is
measured between concepts or properties, each similarity measurement is combined. The
similarity measurements used in ontology matching are lexical similarity, structural similarity,
instance similarity, and logical inference similarity [8, 9].
First, the lexical similarity measure has predominantly two approaches 1) character comparison
through method of words used, name or description ontology’s elements and 2) the use of
WordNet. In the first approach, compound words and words with special characters or spaces
are split into lexical tokens, and then the character comparison is performed. The other approach
generally utilizes synset in WordNet. Existing works using synset in WordNet are described
further in Section 2.3.
Second, ontology’s concepts involve taxonomies and constraints. A taxonomy defines parent-
child relationships of concepts and properties. In ontology matching, similarities between
concepts can be calculated by analyzing ontologies’ taxonomies and various constraints defined
in association with concepts. This approach of comparing taxonomies of two ontologies and
constraints in concepts is called the structural similarity measure.
Third, OWL(Web Ontology Language) defines any instance of a concept in RDF (Resource
Description Framework). The instance similarity measures similarity in concepts by comparing
instances of these concepts.
Fourth, OWL defines property characteristics such as FunctionalProperty, InverseOf,
SymmetricProperty, TransitiveProperty, and InverseFunctionalProperty. Applications can use
property characteristics to make logical inferences, and the reasoned logical inferences are then
used to discover matched concepts. This approach is called the logical inference similarity
measure [9].

4
2.3. Ontology Matching Using WordNet’s synset
WordNet is a thesaurus with defined semantic vocabulary set. Bundles of the same semantic
words are called synset, meaning synonym set and are identified by their synsetID. Synsets of
WordNet are connected with semantic relationships. For example, nouns are linked in terms of
hypernym, hyponym, coordinate term, holonym, and meronym relationship; verbs have the
relationships of hypernym, troponym, entailment, and coordinate terms [10]. Existing ontology
matching methods using WordNet mostly use synsets in WordNet. It matches by the
equivalentClass relationship between two concepts if two concept’s names are included of the
same synset of WordNet.
Lin proposed a method that measures the similarity between two words and computes their
cosine coefficient using WordNet. The similarity here is computed by the rate of the number of
words in synsets to which the word of interest belongs, to the total number of words in WordNet
[11].
Gouli proposed COMPASS (COncept MaP ASSessment tool) which uses WordNet to give
reflective feedback to learners concerning matching results. If a learner used a hyponym or
meronym of the expected word, the matching would be still considered correct. Otherwise, the
tool recommends using a more general concept [12].
In this paper, following an analysis of previous studies about ontology matching, we define the
Property Unmatched Concept. In the next two sections, we present how the ontology matching
method can discover additional matched concepts by measuring the Super Word Set Similarities
between the Property Unmatched Concepts are also presented in detail.
3. ONTOLOGY MATCHING
3.1. Ontology Matching Process
Figure 1. Ontology matching process
As shown in Figure 1, two ontologies, represented as O1 and O2, are needed for ontology
matching. Based on input ontologies O1 and O2, an ontology matching tool first performs
concept matching by measuring the similarity between concepts. Next, property matching is
carried out by measuring the similarity between properties. The logical similarity measure is
then performed based on the results of previous concept matching and property matching.
Lastly, the final matching results are produced after passing through the refinement process.
LOM [6] proposes a lexical similarity measure in ontology matching. LOM performs a lexical
similarity by 4 phases. In the first phase, whole term matching, LOM finds an exact name string
match. In second phase, word constituent matching, stop words such as ‘a’, ‘the’, ‘of’, ‘in’, etc.
are dropped from multi-word terms. Remaining words for each term are compared through
exact string matching. After performing this phase, ‘meeting-place’ and ‘place-of-meeting’ can
be matched. In third phase, Synset matching, if two words are included in synset in WordNet,
two words are matched. Through this phase, ‘auto-care’ and ‘car-maintenance’ can be matched.
In fourth phase, Type matching, LOM performs ontology matching using SUMO(the Suggested

5
Upper Merged Ontology) [13,14] and MILO(the Mid-level Ontology) [15]. Existing ontology
matching tools using a lexical similarity perform first three phases.
Figure 2. Movie ontologies O 1, O2 and matching results. Dotted lines mean a matching
OWL distinguishes between two main categories of properties, datatype properties and object
properties. Datatype properties declare data type defined in XML Schema. Object properties
define semantic relations between concepts. In two concepts semantically linked via an object
property, one becomes a domain concept when the other is a range concept [16]. Object
properties are mainly used in this paper. An object property is represented below.
Ontology_name.(Domain concept_name, Property_name, Range concept_name)
For example, movie ontology O1 in Figure 2 has a property represented as
O1.(Movie,director,Person) where director is an object property, Movie is a domain concept,
and Person is a range concept.
Figure 2 shows the final matching results of two movie ontologies O1 and O2, produced through
the matching processes in Figure 1. The matched concepts O1.Movie and O2.Movie are found in
the phase of similarity measure between concepts. In the phase of similarity measure between
properties, five pairs of properties are matched including O1.(Movie,director,Person) and
O2.(Director,directs,Movie). Table 1 lists the matching results of the two movie ontologies
depicted in Figure 2.

6
Table 1. Matching results of movie ontologies O1 and O2
Analyzing the concepts appearing in the properties O1.(Movie,director,Person) and
O2.(Director,directs,Movie) shows that the domain concept O1.Movie in the property
O1.(Movie,director,Person) and the range concept O2.Movie in the property
O2.(Director,directs,Movie) have been already matched in concept matching phase.
The range concept O1.Person in the property O1.(Movie,director,Person) and the domain
concept O2.Director in the property O2.(Director,directs,Movie) are not yet matched. Let’s
analyze O1.Person and O2.Director. The synset in WordNet of ‘person’ word contains {person,
individual, someone, somebody, mortal, soul} and the synset in WordNet of ‘director’ word
contains {director, manager, managing director}. O1.Person and O2.Director are not matched
because these two words do not contain the same synset of WordNet. In this paper, we propose
an ontology matching approach in order to search new concept matching using hypernyms,
hyponyms, holonyms, and meronyms in WordNet along with synset in WordNet between
O1.Person and O2.Director.
Figure 3. An ontology matching diagram. O1 and O2 ontology examples and matching results.
Dotted lines mean a matching
3.2. Terminology

7
In this paper, a concept matched in the phase of similarity measure between concepts is called a
Matched Concept(MC), represented as O.C. Whereas, O.C represents an unmatched concept.
Similarly, a property matched in the phase of similarity measure between properties is called a
Matched Property(MP), which is denoted as O.(DC,p,RC). An unmatched property is
represented as O.(DC,p,RC). Note that MPs apply only to object properties of the MCs. In
Figure 3, O1.C3 and O2.C2 are matched by means of the relationship equivalentClass, and are
denoted as O1.C3 and O2.C2. There are three pairs of the MPs corresponded by means of the
relationship equivalentProperty, i.e., O1.(C3,p5,C4) and O2.(C2,p1,C3), O1.(C4,p4,C3) and
O2.(C5,p3,C2), and O1.(C3,p3,C4) and O2.(C2,p2,C4). According to the notation described
above, the first pair is denoted as O1.(C3,p5,C4) and O2.(C2,p1,C3).
Among the domain concepts and range concepts of the MPs, those that are not matched are
defined as a Property Unmatched Concept(PUC), and denoted as O.C. In Figure 3, the PUCs
are O1.C4, O2.C3, O2.C4, and O2.C5. They are denoted as O1.C4, O2.C3, O2.C4, and O2.C5.
Figure 4. Example of WordNet’s synonym sets and semantic relationships
WordNet groups words into sets of synonyms, and provides a short, general description of each
set of synonyms. Figure 4.a shows that the word ‘person’ belongs to three synonyms sets. The
first synonym set contains the words {person, individual, someone, somebody, mortal, soul}. In
addition, words found in synset of WordNet are interlinked by various semantic relationships.
For example, WordNet’s nouns are interlinked in terms of hypernym, hyponym, coordinate term,
holonym, and meronym, as listed in Figure 4.b.
This paper defines the Super Word Set that contains all the words and their descriptions in
association with WordNet’s semantic relationships. Note that WordNet’s synsets are not
included in the Super Word Set because they are already used for matching in the phase of
similarity measure between concepts.
SWS(Super Word Set) = ∑ w∈{hyperset}+ ∑ w∈{hyposet}+
∑ w∈{holoset} + ∑ w∈{meroset}
In the equation above, hyperset is a set of hypernyms; hyposet is a set of hyponyms; holoset is a
set of holonyms; and meroset is a set of meronyms. Each of these sets is composed of a set of
words and their descriptions.
4.SUPER WORD SET SIMILARITY
4.1.Extraction of the Property Unmatched Concept

8
The MCs and MPs should be determined first in order to search the PUCs. In the matching
results of the movie ontologies presented in Figure 2 and Table 1, O1.Movie and O2.Movie are
the MCs. Among all the object properties of the MCs, there are five pairs of MPs as shown
below:
O1.(Movie,director,Person) and O2.(Director,directs,Movie),
O1.(Movie,cast,Person) and O2.(Movie,has_casting,Cast),
O1.(Movie,producer,Person) and O2.(Producer,produces,Movie),
O1.(Movie,writer,Person) and O2.(Writer,writes,Movie),
O1.(Movie,editor,Person) and O2.(Editor,edits,Movie).
Then, among the domain concepts and the range concepts of these five pairs of MPs, those that
are not previously matched become candidates for a PUC. To determine whether or not these
candidates should become a real PUC, the property matching rate of each MC is calculated by
dividing the number of MPs by the number of all existing object properties of the MC. If the
calculated property matching rate is greater than a given critical value α (α ∈ [0, 100]), the
candidates are chosen as a PUC.
In Figure 2, the MC O1.Movie has 5 MPs and its total number of object properties is 9. Thus,
O1.Movie has the property matching rate 55% (5/9*100= 55%). The MC O2.Movie has 5 MPs
and 11 object properties, therefore its property matching rate is 45% (5/11*100 = 45%). When
the critical value is 0.45 (α = 0.45), the candidates of O1.Movie and O2.Movie are chosen as a
PUC, i.e., O1.Person, O2.Writer, O2.Editor, O2.Director, O2.Producer, and O2.Cast.
4.2. Super Word Set Similarity Measure
The Super Word Set Similarity is measured to discover new matched concepts based on PUCs.
The Super Word Set Similarity measure proposed in this paper makes use of the Super Word Set
presented in Section 3.2 and WordNet’s synsets. The Super Word Set Similarity measure of two
ontologies Oi and Oj is defined by a percentage of included synset words of the PUC of Oi in the
PUC’s Super Word Set of Oi. The Super Word Set Similarity measure can be computed in two
ways, i.e., from Oi ontology to Oj ontology, and vice versa.
where Oi.C1 and Oj.C2 are a pair of PUCs in two ontologies.
Table 2. A result table of the Super Word Set Similarity of the Property Unmatched Concept

9
Table 2 shows the results of measuring the Super Word Set Similarity between the identified
PUCs of the movie ontologies in Figure 2 - i.e., O1.Person, O2.Writer, O2.Editor, O2.Director,
O2.Producer, and O2.Cast.
The range of the calculated Super Word Set Similarity values in Table 2 is rather broad, from
0% to 44%. Thus, a critical value for the Super Word Set Similarity is used. If the calculated
Super Word Set Similarity of a pair of PUCs is greater than a given critical value β (β ∈ [0,
100]), that pair of PUCs is matched, and the PUCs become new MCs. Based on experiments,
the critical value β is set to 20% (β = 20%). In Table 2, among 10 pairs of PUCs, 3 pairs have
the Super Word Set Similarity measure greater than the given critical value; they are
simsws(O2.Writer, O1.Person), simsws(O2.Director, O1.Person), and simsws(O2.Producer,
O1.Person). These PUCs are thus determined as new MCs.
4.3. Determination of Semantic Relationship
The semantic relationship of a pair of MCs newly discovered by the Super Word Set Similarity
measure also needs to be determined. A critical value β is defined to determine the types of
semantic mapping between new MCs. Based on experiments, γ is set to 80%(γ=80%). The logic
to determine the semantic relationship type based on γ is shown below.
where Oi.C1 and Oj.C2 are a pair of new MCs in two ontologies.
In Table 2, simsws(O2.Writer, O1. Person) has the Super Word Set Similarity measure 33% (less
than the critical value 80%), therefore the pair’s semantic relationship type would be subClass,
denoted as O2.(Writer,subClass,Person).
Figure 5 shows the ontology matching process in which the phase of Super Word Set Similarity
measure is added. The added phase is placed in between the phase of similarity measure
between properties and the phase of logical inference similarity measure.
Figure 5. Ontology matching process added the phase of Super Word Set Similarity measure
5. EXPERIMENTAL EVALUATION
This section explains the metrics used to assess the accuracy and efficiency of ontology
matching, and compares the effectiveness of the proposed method with that of COMA++ and
LOM. Ontologies in the domain of movie and food are used in the experiments.

10
5.1. Measures for Experimental Evaluation
In this paper, the Precision and Recall are used to evaluate the proposed method.
Figure 6. Precision and Recall
This section explains the metrics used to assess the accuracy and efficiency of ontology
matching, and compares the effectiveness of the proposed method with that of COMA++ and
In Figure 6, Ma indicates a set of MCs and MPs via an ontology mapping tool, which is called
automated matching. However, Mm is a set of MCs and MPs that are matched manually by a
domain expert, and is therefore called manual matching. The formula to compute Precision and
Recall are shown below.
Precision is the ratio of the intersection of an automatically matched set and a manually matched
set to the automatically matched set.
Recall is the ratio of the intersection of an automatically matched set and a manually matched
set to the manually matched set.
5.2. Experimental results
In the experiments of the proposed method, the PUCs were identified based on the MCs and
MPs matched in the phase of concept matching and the phase of property matching. The Super
Word Set Similarity was then measured on the extracted PUCs, and new MCs were determined
according to the measured Super Word Set Similarity.
Figure 7 shows the final matching results of two food ontologies O1 and O2, produced through
the matching processes in Figure 1. As shown in Figure 7, O1.Meal and O2.MEAL,
O1.MealCourse and O2.MEAL-COURSE are extracted as the MCs and
O1.(Meal,course,MealCourse) and O2.(MEAL,COURSE,MEAL-COURSE),
O1.(MealCourse, hasDrink,PotableLiquid) and O2.(MEAL-COURSE,DRINK,JUICE) or ,
O2.(MEAL-COURSE,DRINK,WINE) are extracted as the MPs. Based on those matching
results, we can extract O1.PotableLiquid, O2.JUICE and O2.WINE as PUCs.

11
Figure 7. Food ontologies O 1, O2 and matching results. Dotted lines mean a matching.
Table 3. The number of concepts, properties and results of manual matching
Table 3 shows the number of concepts, properties and manual matching of the movie and food
ontologies used in the experiments. Movie ontology O1 has properties eight times higher than
the number of concepts, which indicates that it is property oriented ontology. On the other hand,
movie ontology O2 has the number of concepts more than three times higher than the number of
properties, so it is presumed to be a concept oriented ontology. Food ontologies O1 and O2 are
both concept oriented ontology. Considering that food ontology O2 has the number of concepts
more than two times higher than food ontology O1, one can Figure out that O2 is built in a more
detailed way than O1.

12
Figure 8. Experimental result
In the studies, the proposed method based on the Super Word Set Similarity is compared with
COMA++ and LOM in order to examine the accuracy and efficiency of their ontology matching.
Figure 8 shows the Precision and Recall rates of the two compared methods with regard to
movie ontologies, food ontologies, and a combination of movie and food ontologies.
In the movie ontologies, both Precision and Recall rates are under 60% as the property oriented
ontology O1 and the concept oriented ontology O2 are matched. On the other hand, in food
ontologies O1 and O2 are concept oriented ontology, high Precision and Recall rates are attained
and are over 70%.
The proposed Super Word Set Similarity method has the advantage of finding new MCs based
on the PUCs that are derived from the previously matched concepts and properties. As shown in
Figure 8, the proposed Super Word Set Similarity method performs better than COMA++ in
movie ontology matching. This is because many PUCs were extracted from the MCs and MPs
that are found in the phases of concept matching and property matching. This implies that PUCs
are extracted most effectively when a concept oriented ontology and property oriented ontology
are matched.
On the other hand, the proposed method and COMA++ yield similar results when two concept
oriented ontologies are matched, as shown in the matching results of the food ontologies in
Figure 8. This is because the number of PUCs extracted from the previously MCs and MPs was
relatively small.
6. CONCLUSIONS AND FUTURE WORKS
This paper has presented the Super Word Set Similarity measure that makes use of WordNet to
discover new MCs in ontology matching. The proposed ontology matching method based on the
Super Word Set Similarity takes the matching results of the phase of similarity measure between
concepts and the phase of similarity measure between properties in order to find new MCs. The
experiments carried out demonstrate that the proposed method result in an average improvement
of 12% over COMA++ and 19% over LOM.
Unlike previous ontology matching tools that utilize only WordNet’s synonym sets, the
proposed method employs semantic relationships set of WordNet - i.e., Super Word Set
including collectively WordNet’s hypernyms, hyponyms, holonyms, and meronyms is used for
ontology matching. In the proposed method, additional MCs are newly discovered based on the
inclusion rate of synset words over Super Word Set. Currently, the proposed Super Word Set
Similarity method is applied for concepts that have a one-word noun. In future work, how to
address compound nouns, adjectives and verbs will be studied. That is, this research will be

13
extended to build a Super Word Set that includes all types of words so that the proposed method
is an applicable to a wider range of word types.
ACKNOWLEDGEMENTS
This work was supported by National Research Foundation of Korea Grant funded by the
Korean Government(2009-0077026).
REFERENCES
[1] AnHai Doan, Pedro Domingos, and Alon Halevy, (2003) “Learning to Match the Schemas of
Data Sources: A Multistrategy Approach,” Machine Learning, Vol. 50, No.3, pp 279-301.
[2] Namyoun Choi, Il-Yeol Song, and Hyoi Han, (2006) “A Survey on Ontology Mapping,” ACM
SIGMOD Record, Vol. 35, No, 3, pp 34-41.
[3] AnHai Doan, Jayant Madhavan, Robin Dhamankar, Pedro Domingos, and Alon Y. Halevy,
(2003) “Learning to match ontologies on the Semantic Web,” The VLDB Journal, Vol. 12, No 4,
pp 303-319.
[4] Domenico Beneventano, Sonia Bergamaschi, Francesco Guerra, and Maurizio, (2003)
“Synthesizing an Integrated Ontology,” IEEE Internet Computing, pp 42-51.
[5] Alexander Maedche, Boris Motik, Nuno Silva, and Raphael Volz, (2002) “MAFRA - A
MApping FRAmework for Distributed Ontologies in the semantic Web,” Proc. of the Workshop
on Knowledge Transformation for the Semantic Web (KTSW 2002), pp 60-68.
[6] John Li, (2004) “LOM: A Lexicon-based Ontology Mapping Tool,” Proc. of the Performance
Metrics for Intelligent Systems Workshop (PerMIS. ’04), pp 1-5.
[7] David Aumueller, Hong-Hai Do, Sabine Massmann, and Erhard Rahm, (2005) “Schema and
ontology matching with COMA++ ,“ Proc. of the ACM SIGMOD International Conference on
Management of Data, pp 906-908.
[8] Jie Tang, Juan-Zi Li, Bangyong Liang, Xiaotong Huang, Yi Li and Kehong Wang, (2006)
“Using Bayesian decision for ontology mapping,” Journal of Web Semantics, Vol. 4, No. 4, pp
243-262.
[9] Octavian Udrea, Lise Getoor, and Renée J. Miller, (2007) “Leveraging data and structure in
ontology integration,” Proc. of the ACM SIGMOD International Conference on Management of
Data, pp 449-460.
[10] WordNet, http://wordnet.princeton.edu/
[11] Patrick Pantel and Dekang Lin, (2002) “Discovering Word Senses from Text,” Proc. of the
Eighth ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pp
613-619.
[12] Harry Kornilakis, Maria Grigoriadou, Kyparisia A.Papanikolaou, and Evangelia Gouli, (2004)
“Using WordNet to Support Interactive Concept Map Construction,” Proc. of the IEEE
International Conference on Advanced Learning Technologies (ICALT’04), pp 600-604.
[13] Ian Niles and Adam Pease, (2001) “Toward a Standard Upper Ontology,” Proc. of the
International Conference on Formal Ontology in Information Systems (FOIS-2001), pp 2-9.
[14] Adam Pease, and Ian Niles, (2003) “Linking Lexicons and Ontologies: Mapping WordNet to the
Suggested Upper Merged Ontology,” Proc. of the IEEE International Conference on Information
and Knowledge Engineering, pp 412-416.
[15] Ian Niles, and Allan Terry, (2004) “The MILO: A general purpose, mid-level ontology,” Proc. of
the International Conference on Information and Knowledge Engineering (IKE'04), pp 15-19.
[16] Grigoris Antoniou and Frank van Harmelen, (2004) A Semantic Web Primer, The MIT Press,
Massachusetts, pp 118.

14
Authors
JungAe Kwak is a Ph.D. candidate at Computer Science
and Engineering at Ewha Womans University, Republic of
Korea. She received the M.S. degree in Information Science
and Engineering at Ewha Womans University. Her current
major research interests include semantic web, ontology
matching, and ontology modelling.
Hwan-Seung Yong is a Professor of Computer Science and
Engineering at Ewha Womans University, Republic of
Korea since 1995. He received the B.S., M.S. and Ph.D.
degrees in Computer Engineering from Seoul National
University. He has five years of industrial experience as
Researcher at ETRI (Electronics and Telecommunications
Research Institute) and served as a visiting scientist at the
IBM T.J. Watson Research Center. He is member of KIISE
(The Korean Institute of Information Scientists and
Engineers) and ACM. His current major research interests
include agile methods, data mining, multimedia database
and ontology search.
Short Biography

Ontology Matching Based on hypernym, hyponym, holonym, and meronym Sets in WordNet

More Related Content

What's hot (19)

Similar to Ontology Matching Based on hypernym, hyponym, holonym, and meronym Sets in WordNet (20)

More from dannyijwest (20)

Recently uploaded (20)

Ontology Matching Based on hypernym, hyponym, holonym, and meronym Sets in WordNet