International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

SEMI AUTOMATIC TO IMPROVE ONTOLOGY MAPPING PROCESS IN

SEMANTIC WEB DATA ANALYSIS
*1Ms. Suganya E., *2 Ms. Abarna N

*1M.Phil Research Scholar, PG & Research Department of Computer ScienceArcot Sri Mahalakshmi Womens
College , Vilapakkam, Tamil Nadu, India
*2.Assistant Professor, PG & Research Department of Computer Science Arcot Sri Mahalakshmi Womens College,

Vilapakkam, Tamil Nadu, India

---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract: Ontology Mapping is an important issue, widely agents) would be able to automatically find and process
recognized in a research community of Semantic Web. knowledge[8]. The Semantic Web, as the name implies, is a
Large number of ontologies, developed across the World Web focused on the conveyance of meaning. It facilitates
Wide Web in a distributed manner demands for automatic modeling, sharing and reasoning with knowledge available
or at least semi-automatic ontology mapping system to on the Web through the formal representation of
integrate information from different ontologies and to make knowledge domains with ontologies (roughly, agreed
the vision of Semantic Web reality. The integrated approach terms)[8].The present HTML based WWW defines Web
here means two things: (1) it integrates several techniques pages syntactically and were intended only for human
from different computational area such as Computational consumption. The HTML defines how content of the Web
Linguistic, Information Retrieval, and Machine Learning in pages should be displayed on browser, but does not tell
order to provide semi-automatic ontology mapping process; anything about the subject and nature of the content.
and (2) It takes care of ontology mapping process right Hence, these Web pages cannot be read and processed by
from the creation of the ontologies. The algorithm performs machines without human intervention to derive any
these steps in iterative and interleaved manner depending meaning out of it. With the Semantic Web, information on
on its execution configuration. The language processing the Web can be defined semantically in such a way that it
activities such as Tokenization, Lemmatization, can be used by machines, not only for display purposes,
Abbreviation Expansion, Spelling Correction, Elimination of but also for interoperability and integration [4].
Stop words, etc. are performed whenever required. It uses
domain specific thesaurus for abbreviation expansion and 1.2 Semantic Web Layered Architecture
synonym. It also uses Word Net, an online lexical database,
to strengthen the Linguistic Matcher. The AI - ATOM has The Semantic Web is consisting of a philosophy
been evaluated using the measures precision, recall, and F- (idea), a set of design principles, collaborative working
measure on two small real world data sets. The system is groups, and a variety of enabling technologies. Some
tested with different algorithm configuration to decide the elements of the Semantic Web are expressed as
best possible default configuration for the domain under prospective future possibilities that have yet to be
consideration. The preliminary case studies show implemented or realized. Other elements of the Semantic
encouraging result. Web are expressed in formal specifications [16]. Some of
these formal specifications include Resource Description
Keywords: Ontology Mapping, ATOM, wordnet, Framework (RDF), a variety of data interchange formats
Tokenization, Semantic Web, F-Measure Algorithm, (e.g., RDF/XML, N3, Turtle, and N-Triples), and notations
XML/ RDF Framework. such as RDF Schema (RDFS) and the Web Ontology
Language (OWL), all of which are intended to provide a
I. INTRODUCTION formal description of concepts, terms, and relationships
within a given knowledge domain. Key technologies in
Semantic Web include explicit meta-data, ontologies, logic
1.1 Semantic Web and inference, and intelligent agents [7].
The present WWW has huge collection of pages,
but majority of them are in human readable format only.
As a consequences software agent cannot understand and
process this information, and much of the potential of the
Web has so far remain untapped. To overcome this
problem, delivered at a meeting of the World Wide Web
Consortium (W3C), Tim Berners-Lee presented his vision
for a new Web, Semantic Web, where machines (software

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 183
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

Trust Layer include the concept of digital signature along

with other means such as certificate and trusted agencies
as similar to the one, already implemented in present Web.

Linguistic element level matching: It uses the

information such as names, data types, and domains of
element. It also uses tokenization for compound element
names before computing element level score. For name
matching it uses morphological normalization,
categorization, string based techniques such as common
prefix and suffix; and a thesauri lookup[12].

Structural matching: This component transforms input

schemas into trees that enrich the structure by
augmenting referential constraints. The linguistic
similarity of node and similarity of their leaf nodes are
Fig.1.1 Semantic Web Layered Architecture combined to decide similarity between two elements.
Finally, it calculates weighted mean of linguistic and
This diagram shows set of technologies, tools, and structural similarity.
standards; organized into a certain structure that is an
expression of their interrelationships; which form the Mapping identification: This step depends on the
basic building blocks of an infrastructure to support the application in which the algorithm is used and decides the
vision of the Web associated with meaning[6]. mappings based on weighted mean calculated in previous
phase [2][1].
3-step methodology for knowledge integration, which
consists in: Cupid represents the input schemas internally as
trees. These trees are processed by name matchers
Expressing ontological content in a linguistically assisted with auxiliary thesaurus to get linguistic
motivated fashion, as a necessary part of the similarity at element level and then by structural matchers
development of ontologies that uses name and data type information of leaves in an
Automatically discovering linguistic and semantic iterative manner with knowledge propagation until the
evidences to suggest conceptual similarities desired threshold is exceeded. It combines linguistic and
during automatic ontology alignment. structural similarities using a weighted sum. Finally, it
Supporting users in the process of producing generates the alignments based on this weighted
assessed ontology mapping documents, offering similarity crossing some threshold values[7][8].
reliable knowledge for providing semantic links
across different information sources II. RELATED WORK

2.1 ONTOLOGY MAPPIN SYSTEM

COMA

COMA represents the schema internally as a

directed acyclic graph where elements are the paths. This
is done to capture the contexts in which elements occur. It
allows the user interaction to improve the accuracy of
matching based on approval of suggested matches or
mismatches. It may also be used to evaluate the different
combinations of matchers. The main components of COMA
are: the Repositoryto persistently store all match-related
Fig.1.2 Modified Semantic Web Layered Architecture data, the Modeland Mapping Poolsto manage schemas,
ontologies and mappings in memory, the Match
This diagram represent slightly modified diagram Customizerto configure matchers and match strategies,
for Semantic Web Layered Architecture. There are two and the Execution Engineto perform match operations.
reasons considered for this modification: first, OWL, a new The latest version of COMA, COMA++, improves the
recommendation from W3C for writing ontology, is algorithm and provides a GUI. The Figure 16 shows the
missing in original diagram; and second, I believe that overall architecture of COMA++[8].

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 184
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

Fig2.1 Architecture of COMA++

2.2 Cupid

Cupid (Microsoft Research) implements a hybrid matching

algorithm combining linguistic and structural schema
matching techniques. It calculates the normalized
similarity with the help of external thesaurus. It is generic Fig2.2: GLUE Architecture
in the sense that it can be applied to XML as well as
relational schemas.Cupid represents the input schemas III. PROPOSED ANALYSIS
internally as trees[2]. These trees are processed by name
matchers assisted with auxiliary thesaurus to get linguistic 3.1 Terminology
similarity at element level and then by structural matchers
that uses name and data type information of leaves in an Ontology: Ontology is a 6-tuple denoted by O = {C, CH, P,
iterative manner with knowledge propagation until the PH, I, A}, where C is the set of concepts (Classes), CH
desired threshold is exceeded. It combines linguistic and defines Class Hierarchy (the hierarchical relationships
structural similarities using a weighted sum. Finally, it among the concepts), P is the set of properties (Object
generates the alignments based on this weighted Property defining the relationship between concepts and
similarity crossing some threshold value[11]. Data Property defining the attributes of the concepts), PH
defines Property Hierarchy (the hierarchical relationships
2.3 GLUE among the properties), A is the set of axioms, I is the set of
instances of concepts and properties. The present work
GLUE uses machine learning technique to find considers only C and CH components of ontology.
mappings. It employs multiple learning strategies and uses
different types of knowledge to improve the process. GLUE Ontology Mapping: Ontology mapping finds
finds the most similar concepts between two ontologies correspondences between semantically related entities of
and calculates the joint probability distribution of the ontologies covering the same domain in such a way that
concept using a multi-strategy learning approach for intended interpretations in shared domain are
similarity measurement. It uses machine learning preserved[3][9].
techniques to compute for every pair of concepts their
joint probability distribution. For this purpose it Mapping Element: A mapping element (also known as
implements two base learners, viz., Content Learner and correspondence) is a 6-tuple: ME=(Id, e1, e2, n, r, uid),
Name Learner; and third learner called Meta Learner that where Id is a unique identifier of the given
combines the two base learners' prediction. But, it correspondence; e1 and e2 are entities (e.g., Class, Object
involves huge manual effort to train these algorithms Property, Data Property, or Instance) of the first and the
before they can be used effectively. It also uses a technique second ontology, respectively; n is a confidence measure
called Relaxation Labeling that assigns labels to nodes of a (typically in the [0, 1] range) holding for the
graph, given a set of constraints. It consists of three correspondence between e1 and e2; and r is a relation
modules, viz.,Distribution Estimator, Similarity Estimator, holding between e1 and e2, which could be, for example,
and Relaxation Labeler. The Figure 19 shows the overall equal (=),more general (>), less general (<), or disjoints
architecture of GLUE[16]. (!); and uid is ID of user who approved this mapping

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 185
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

element. The present work considers only equality While user creates new ontology, the system
relation. suggests the concept labels along with their meaning from
the user specified context dictionary (Domain Specific
Concept's Local Label: Concept's Local Label is the class Labels), based on its usage frequency, that are best
name given by user to describe the concept. matching with partially entered labels by user. The context
dictionary is maintained with the help of following table.
Concept's Path Label: Concept's Path Label is Label Domain Specific Label (DSL ID, Label, Meaning, Frequency
derived based on path from root (or grandparent or of Use).
specified super class) node to class node. It augments
labels of all parent class nodes with label of class node.

The basic objective of the proposed system is to

increase the automation in ontology mapping process by
using all possible source of information from ontology and
various techniques in an integrated manner[12].

A GUI based system AI-ATOM is developed as a prototype

for the proposed integrated approach. The system consists
of five horizontal components, viz., Ontology Management,
Ontology Project Management, User Management, System
Configuration and Ontology Mapping Engine; and six
vertical components, viz., Language Processing, VSM
Engine, Label Matcher, Linguistic Matcher, Structure
Matcher, and User Interaction[5]. The brief description of
each component is given below. It allows the user to create
new ontology, maintain an existing ontology, delete
existing ontology, and to import ontology from file
containing parenthesized tree. While user creates new
ontology, the system suggests the concept labels along
with their meaning from the user specified context
dictionary (Domain Specific Labels), based on its usage Fig3.2: General Tree Representation using parenthesis
frequency, that are best matching with partially entered for Ontology
labels by user. The context dictionary is maintained with
the help oftable. This helps the user from typing and This helps the user from typing and thinking
thinking different label from intention of the community of different label from intention of the community of users in
users in the same domain. This helps a lot when two the same domain. This helps a lot when two such
suchontologies need to be mapped later. Thus, ontologies need to be mapped later. Thus, system
systemproposed here tries to improve ontology mapping proposed here tries to improve ontology mapping process
process right from its creation. right from its creation.

3.2 Data Structure

The Data Structure used by Algorithm is listed below:

User (User ID, User Name, Password)

Ontology (Ontology ID, Ontology Name,
Description, Owner User ID, Date of Creation)
Class (Class ID, Class Name, Description, Ontology
ID
Class Has Sub Class (CHSC ID, Class ID, Sub Class
ID)
Ontology Mapping Project (OMP ID, OMP Name,
Description, From Ontology ID, To Ontology ID,
Created By User ID, Start Date, Due Date,
Completed Date)
Ontology Mapping Project User (OMPU ID, OMP
ID, User ID)
Fig3.1: Architecture of AI-ATOM

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 186
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

Domain Specific Label (DSL ID, Label, Meaning, End If

Frequency of Use) End If
Domain Specific Abbreviation (DSA ID, Label Matcher Similarity=Max (Label Matcher
Abbreviation, Expanded Text) Similarity, Edit Distance Similarity)
Domain Specific Synonym (DSS ID, Synonym, If (Label Matcher Threshold Should Be Used)
Synonym Group ID) Then
Stop word (Stop word ID, Stop word) If (Label Matcher Similarity >= Label Matcher
Algorithm Configuration (AC ID, AC Name, Threshold) Then
Description, For OMP ID, Precision, Recall, F SELECT ME(Ci,Cj)
Measure) Process Next ME (Cp, Cq)
Parameter (Parameter ID, Parameter Name, End If
Description, Default Value Text, Default Value End If
Number) End If
Parameter Possible Value (PPV ID, Parameter ID,
Possible Value Text, Possible Value Number) //Compute N-Gram
Algorithm Parameter Value (APV ID, For AC ID, IF (NGram Should Be Used) Then
Parameter ID, User PPV ID) NGram Similarity=get NGramScore(Ci, Cj)
If (NGram Threshold Should Be Used) Then
Mapping Element (ME ID, OMP ID, From Class ID,
From Class Name, To Class ID, To Class Name, If (NGram Similarity >= NGram Threshold) Then
Status, Relation, Confidence, Explanation, SELECT ME(Ci,Cj)
Process Next ME (Cp, Cq)
Processed By System YN, Processed By User YN,
User ID) End If
End If
Status=Candidate/Potential/System
Label Matcher Similarity=Max (Label Matcher
Generated/Accepted/Rejected
Similarity, NGram Similarity)
Accepted Rejected Mapping Group (ARMG ID,
If (Label Matcher Threshold Should Be Used) Then
Class Name, Relation, Group ID, Accepted or
If (Label Matcher Similarity >= Label Matcher
Rejected).
Threshold) Then
SELECT ME(Ci,Cj)
Process Next ME (Cp, Cq)
3.3. Ontology Mapping Project Management
End If
End If
It allows the users to create, maintain, and delete
End If
an ontology mapping project. It also allows assigning users
/*
to mapping project, and to review completed and pending
mapping pairs which are yet to be processed for its
Similarly, perform following Label sub-matchers.
acceptance or rejection [3]. Thus, this system supports a
new approach of multi-user, multi-session ontology
i. N-Gram
project management in order to reduce burden on
ii. Prefix and Suffix
singledomain expert and to provide the time flexibility to
iii. Soundex
users.
iv. Heuristic Rules-1 to N
If (Label Matcher Threshold Should Be Used) Then
Mapping Algorithm If (Label Matcher Similarity >= Label Matcher Threshold)
Then
Linguistic Matcher Similarity=0.0 SELECT ME (Ci,Cj)
Structure Matcher Similarity=0.0 Process Next ME (Cp, Cq)
System Similarity=0.0 End If
IF (Label Matcher Should Be Used) Then System Similarity = Max (System Similarity, Label
Matcher Similarity)
//Compute Edit Distance If (System Similarity Threshold Should Be Used)
IF (Edit Distance Should Be Used) Then If (System Similarity >= System Similarity
Edit Distance Similarity=get Edit Distance Score (Ci, Cj) Threshold) Then
If (Edit Distance Threshold Should Be Used) Then SELECT ME(Ci, Cj)
If (Edit Distance Similarity >= Edit Distance Process Next ME(Cp, Cq)
Threshold) Then End If
SELECT ME(Ci,Cj) End If
Process Next ME (Cp, Cq) End If

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 187
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

If (Linguistic Matcher Should Be Used) Then

If (Domain Synonym Should Be Used) Then
If (isDomainSynonym (Ci, Cj)) Then
SELECT ME(Ci, Cj)
Process Next ME(Cp, Cq)
End If
End If
If (WordNet Synonym Should Be Used) Then
If (isWordNetSynonym(Ci, Cj)) Then
SELECT ME(Ci, Cj)
Process Next ME(Cp, Cq)
End If
End If Fig3.3 : Precision, Recall, and F-Measure
If (WordNet Gloss Should Be Used) Then
WordNet Gloss Similarity = getWordNetGlossScore(Ci, Cj) Using these notions; Precision, Recall, and F-Measure can
System Similarity=System Similarity + (1- be defined as following.
System Similarity) * WordNet
Gloss Similarity Recall = | Relevant and Retrieved | / | Relevant |
End if = B / (B+C).
End If Precision = | Relevant and Retrieved | / |
Retrieved |= B / (A+B).
IV. EVALUATION RESULT F-Measure= 2 * (Precision * Recall) / (Precision +
Recall) = 2 * B/ ((A+B) + (B+C)).
The effectiveness of the system in processing the
mapping elements is measured by looking at Precision and The first data set represents sample ontologies
Recall [67] [24]. The ontology mapping systems are represented by two different academic institutes, whereas
evaluated with respect to the notion of correctness second data set represents snapshot of database schema
perception a judgment by a human that a mapping from two academic institutes. Both the data sets are
element found by ontology mapping algorithm is correct selected from the academic domain as one of the
or not. A systems ability to retrieve correct mapping objectives of study is to analyze and to present the
elements is assessed with a Recall measure that is defined significance of domain knowledge in the automated
as below: ontology mapping process. In experimental setup, these
Recall = | Relevant and Retrieved | / | Relevant | data sets are given to few users having varying knowledge
regarding the ontology mapping.
A system can achieve 100% recall by simply
returning all the possible mapping elements between two
ontologies.A systems accuracy is based on how many of
the mapping elements generated by system areactually
correct as per users decision, which can beassessed by a
Precision metric and is defined below.
Precision = | Relevant and Retrieved | / | Retrieved |

Ideally both should be 100%. But, most of the

system scarifies one for the other. Hence, to measure the
optimum balance between these two measures, F-Measure
is used which is defined as below:
F-Measure = 2 * (Precision * Recall) / (Precision + Recall)

The Precision and Recall can be understood from

the Figure 47, Where:

A = False Positives
B = True Positives
C = False Negatives Data Set4.1: Sample Ontologies from two different
D = True Negatives Academic Institutes

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 188
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

basic learning component which takes advantage users

feedback to improve ontology mapping process in future.

FUTURE WORK
The VSM is used in algorithm for filtering weak
mapping elements, but it may be improved and used as a
matcher to find mapping elements whose VSM similarity
(cosines score) crosses specified threshold. The advantage
of algorithm configurability with different inclusion of
matchers, their order of execution, and with different
threshold values needs to be assessed. The evaluation of
the algorithm for different combinations of matchers and
Fig4.2: Overall System Performance for Data set system parameters, a gigantic job, is yet to be
System Performance against User performed.The algorithm can be improved using
knowledge of all components of ontology such as instances
and complex relationship.

REFERENCES

1. Abels, S., Haak, L., & Hahn, A. (2005).

Identification of Common Methods Used for
Ontology Integration Tasks. IHIS05. Bremen,
Germany.
2. Ahmed, A., Volker, H., & Nematollaah, S. (2008).
An Empirical Comparison of Ontology Matching
Techniques. Journal of Information Science, 1-20.
3. AMIA2003 Tutorial.ppt. (n.d.). Retrieved 11 2, 12,
fromprotege.stanford.edu/amia2003/AMIA2003T
Fig4.3: System Performance against User for Data set utorial.ppt.
4. Antoniou, G., &Harmelen, F. v. (2008). A Semantic
V.CONCLUSION Web Primer (Vol. 2e). Massachusetts London,
England: The MIT Press Cambridge.
An Integrated approach is proposed that takes 5. Beckett, D., &Broekstra, J. (2008). SPARQL Query
care of ontology mapping process from the very first step Results XML Format. W3C Recommendation.
of ontology creation by allowing the user to set and use World Wide Web Consortium.
domain specific context dictionary and allowing the user 6. Borst, W. N. (1997). Construction of Engineering
to set other domain specific thesaurus such as Ontologies for Knowledge Sharing and Reuse.
abbreviation and synonym to provide context information Enschede, The Netherlands:University of
to ontology mapping process that improves automation of Tweenty.
ontology mapping process. It effectively combines the 7. Choi, N., Song, I.-Y., & Han, H. (2006, September).
power of vector Space Model, used in information A Survey on Ontology Mapping. SIGMOD Record ,
Retrieval, to generate potential candidate mapping 35 (3).
elements to be selected for further processing. Thus it 8. Clark, K. G., Feigenbaum, L., & Torres, E. (2008).
expedites the process by eliminating large number of SPARQL Protocol for RDF. W3C Recommendation.
weak candidate mapping elements. It proposes the novel World Wide Web Consortium.
approach of using previously rejected mappings to speed- 9. def.txt. (n.d.). Retrieved 2004, from
up the process in addition to re-use of accepted mappings. http://www.tartarus.org/:http://www.tartarus.o
rg/~martin/PorterStemmer/def.txt
It supports extensive configuration of algorithm. 10. DERI-TR-2003-10-29.pdf. (n.d.). Retrieved 11 2,
User can select matchers and sub-matchers to be included 12, from www.deri.ie/:
in algorithm and can decide their execution order. Allows www.deri.ie/fileadmin/documents/DERI-TR-
the users to work in multi-user and multi-session 2004-06-30.pdf
environment. A set of heuristic rules with high degree of 11. Do, H.-H., & Rahm, E. (2002). COMA - A System for
feasibility is used in Label Matcher and in Structure Flexible Combination of Schema Matching
Matcher to support automation. The algorithm includes a

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 189
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 08 | Aug -2017 www.irjet.net p-ISSN: 2395-0072

Approaches. 28th Intl.Conference on Very Large

Databases (VLDB). Hong Kong, China.
12. Doan, A., Madhavan, J., Domingos, P., & Halevy, A.
(2003). Learning to Map between Ontologies
onthe Semantic Web. VLDBJournal, Special Issue
on the Semantic Web .
13. Dublin Core Metadata Initiative (DCMI). (n.d.).
Retrieved February 28, 2013, from Dublin Core:
http://dublincore.org/
14. Ehrig, M., &Staab, S. (2004). QOM Quick
Ontology Mapping JA. ISWC 2004, LNCS 3298, (pp.
683-693).
15. Ehrig, M., &Staab, S. (2004). QOM Quick
Ontology Mapping Report. LNCS, Institute AIFB,
University of Karlsruhe
16. Euzenat, J., Isaac, A., Meilicke, C. S.,
Stuckenschmidt, H., Svab, O., Svatek, V., et al.
(2007). First results of the Ontology Alignment
Evaluation Initiative 2007. Second Intrnational
Workshop on Ontology Mapping. IEEE Xplore.
17. Euzenat, J., Loup, D., Touzani, M., &Valtchev, P.
(2004). Ontology alignment with OLA.
Proceedings of the 3rd EON Workshop, 3rd
International Semantic Web Conference.

2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 190