©2010 International Journal of Computer Applications (0975 - 8887)
Volume 1 – No. 26
135
Ontology based Similarity Measure in Document
Ranking
Sridevi.U.K
Senior Lecturer
Department of Applied Sciences
Sri Krishna College of Engg and tech
Coimbatore, India
Nagaveni .N
Assistant Professor
Department of Mathematics
Coimbatore Institute of Technology
Coimbatore, India
ABSTRACT
This paper presents a methodology for the ontology based semantic
annotation of web pages with annotation weighting scheme that
takes advantage of the different relevance of structured document
fields. The retrieval model is based on the importance factors of the
structural elements, which are used to re-rank the documents
retrieval by the ontology based distance measure. The relevance
concept similarity are combined with the annotation-weighting
scheme to improve the relevance measures. The proposed method
has been evaluated on USGS Science directory collection.
Preliminary experiments results show that our method may
generate relevant document in the top rank.
Keywords
Ontology, Information Retrieval, Annotation,
Semantic Search
1. INTRODUCTION
The rapid growth of documents, web pages and other types of textual
content pose a great challenge to modern content management
systems. Ontologies offer an efficient way to reduce the amount of
information overload by encoding the structure of a specific domain
and offering easier access to the information for the users. The degree
to which different Web page elements are indicative of its content is
in this paper referred to as significance indicator. The technique that
determines the importance of different parts of a Web page improves
the retrieval performance. However, all major ontology editors (such
as Protégé[2], OntoStudio[3], are fully manual and offer little support
to the users for structuring domains. Today’s Web search
technologies rely on link analysis techniques that exploit the structure
of the Web to determine the important documents.
Because of the limitations of traditional retrieval mechanisms,
conventional direct keyword based information retrieval technology
cannot meet the growing user retrieval need with semantic
knowledge. The keyword-based retrieval fails to integrate
information spread over different resources. The information retrieval
model does not utilize Semantics of the queries and document
collection [7]. There have been many works which employ the
Semantic Web technologies for information and retrieval such as
TAP [4], KIM [11].A variety of aspects on improving search and
ranking documents have been considered, such as concept based
search of documents [9] [14]
The Semantic Web aims to achieve better data automation,
reuse and interoperability [3]. The main advantage of Semantic Web
is to enhance search mechanisms with the use of Ontology’s [12].
Ontology is a general description of all concepts as well as their
relationship. The Resource Description Framework /Schema
(RDF(S)) and Web Ontology Language (OWL) are W3C
recommended data representation models which are used to represent
the ontology’s [10] [18]. The basic method for constructing the
Semantic Web is to use the terms defined in ontology as metadata to
markup the Web’s content. It is generally accepted that ontology
refers to a formal specification of conceptualization.
In this paper, we propose an ontology based
retrieval model for the exploitation of environmental sciences domain
ontology’s and knowledge bases, to support semantic search in
document repositories. The research problem of improving relevance
in search and ranking of documents requires techniques that consider
the semantic annotation. The search system takes advantage of
ontology based semantic annotation and it includes the weights of
document structure in ranking [2]. Our current work is motivated by
the need of new tools that can improve the retrieval and integration of
information. In this context, we focus on ontologies whose
specification components include entity classes, semantic relations
among these classes, and distinguishing features that describe these
classes. We evaluate our system on the collection of documents from
USGS Science directory. Experimental results indicate that
combining the ontology distance measure and semantic annotation
weights improves the retrieval performance.
2. RELATED WORK
Over the last three years, the numbers of Semantic Web tools have
been developed. The current research focus on providing a semantic
metadata that enhances the information retrieval and support e-
business applications. Automatic creation of metadata for Web pages
resembles the task of semantic annotation in general[17]. A number
of annotation tools for producing Semantic markups exist such as
SHOE, Protégé, OntoAnnotate, MnM [4] [9]. There are several
research projects about ontology-based information retrieval. The
ontology definition of concepts can be used to describe the concepts
and these concepts will be defined as document class [1]. SEAL [9]

©2010 International Journal of Computer Applications (0975 - 8887)
Volume 1 – No. 26
136
was conceived for semantic search of knowledge on the Web and has
also been used for sharing knowledge on the WEB.
Semantic annotation is about assigning to the entities in the text links
to their semantic description [10]. Annotation provides additional
information about Web contents so that better decision on content can
be made. Annotation ontology tells what kind of property and value
types should be used in describing a resource. The usage of domain
ontology’s are employed for the annotations. To improve the
recognition of important indexing terms, it is possible to weight the
concepts of a document in different ways. For example, in topic
indexing, concepts that form semantically related terms, gain more
weights. Although various annotation systems and methods have
been developed, the question of how to easily and cost effectively
produce quality metadata still remains largely unanswered. Dublin
core annotation mainly describes properties of the document itself
without providing too many details about its content. Ontology based
annotations are instead developed to describe the content of the
document and not its general properties. The manual annotation of
document is a high cost and error prone task. To alleviate this task, an
important effort is currently being made in automation of document
annotation and the result is some degree of automation. However
there is still some work to do achieve a complete automation of the
annotation .The classical information retrieval model is incapable of
supporting logical inference.
In general, most of the work about semantic annotation
requires some predefined ontology’s to extract, define and relate the
annotation. The models of automatic semantic annotation are
ontology driven semantic tagging and semantic meta data generation.
The automatic semantic meta data generates Meta data that can
semantically describe the content of annotating the page. The
generated Meta data includes ontology by system defining its own
semantic categories or a system relies on some predefined ontology.
The annotation process of a web page is based annotating a web page
with ontology and adding the relations between individuals. The
ontology-based information retrieval based on vector space model
describes the semantic annotation scheme of KIM platform [9]. It has
reused automatic concept to label mapping available from the
KIMKB [11]. In fully automatic annotation systems like KIM
architectures support instance identification in a restricted predefined
ontology model. Ontology based semantic annotation are needed
when building the Semantic Web. The ontology-based information
retrieval recognizes the relations among terms by referring to the
ontology. Creating ontology’s is not an easy task and obviously there
is no unique correct ontology for any domain. The real quality of
ontology can be assessed only for its use in real application. An
ontology is a type of knowledge base that describes concepts through
definitions that are sufficiently detailed to capturethe semantics of a
domain. An ontology captures a certain view of the world, supports
intentional queries regarding the content of a database, and reflects
the relevance of data by providing a declarative description of
semantic information independent of the data representation [11].
KIM [11] introduces a holistic architecture of Semantic annotation,
indexing and retrieval for documents. It aims to achieve fully
automatic annotation and to improve search and retrieval by
integrating information extraction using GATE. Ontology based
retrieval model work [9] complements KIM with a ranking algorithm
specifically designed for ontology based retrieval model using
semantic indexing scheme based on annotation weighting techniques.
Genetic algorithms are generally quite effective for rapid global
search to find solutions in nondeterministic problems. Genetic
algorithm method enhances the efficiency and adaptability of a meta
searching [15].Research in structural weights has suggested using
document structures for document ranking. Genetic mining of HTML
structures [15] uses the HTML tag weights to improve the
performance of document retrieval system. The term that exists in
title, bold, anchor tag adds more weights to the document then other
terms. The document retrieval performance is improved depending on
the structural importance of the document. Recent investigations in
information retrieval and data integration have emphasized the use of
ontologies and semantic similarity functions as a mechanism for
comparing objects that can be retrieved or integrated across
heterogeneous repositories[5],[6][8][16],[11],[14]. The distribution
frequency of keyword improves the identification of important
document based on the query [9]. The new similarity measure that
incorporates the tag structure weights in addition to standard
weighting schemes. Our approach combines the ontology similarity
distance with annotation scheme to rank the annotated documents.
The ontology based distance approach and annotation scheme
significantly improves the retrieval performance especially for the top
ranked document
3. ONTOLOGY BASED SIMILARITY
3.1 Document Representation
Document is composed of many terms and important words are
spread out documents. The importance of significance indicators
assigned to the Web elements like title, heading, bold, anchor
improves the ranking of the Web documents. Unlike text documents,
Web pages have certain characteristics such as structural information,
hyperlinks and anchors which could serve as potential indicators of
subject content. The relevance score of the document is assigned
based on which term is matched and the part of the Web page in
which the match is found. The annotation process of a web page is
done with concepts of the ontology. Then, relations between
individuals are discovered and instances are added. Document
should be preprocessed to obtain semantic annotations and indexing
of the document should be done. Examine the location of the
annotated instance in the document. The annotation weights are
calculated by combining the frequency and structures weights.
Web search engines provide advanced features in that a user can
specify how a query is matched with title of the page, text of the
page, URL and links to the page, anywhere in the page. Although it is
worthwhile to investigate how different matching affects the accuracy
of the similarity of query to the document of retrieval results using
tag weights [9]. Suppose a document collection on the Semantic Web
is D= {d1, d2…,dn}. The number of occurrences of an instance in a
document is primarily defined as the number of times the label of the
instance appears in the document, if the document is annotated with
instance and zero otherwise.
To extend vector space model to support structured ranking
occurrences within each document structure must be included. The
weight of a term in a document is basically computed by the classical
tf.idf. The term frequency (TF) is the number of times that a term t
appears in a document. The inverse document frequency (IDF) is the
inverse of document frequency in the collection that contains term.
The weighting scheme is defined in (1).

©2010 International Journal of Computer Applications (0975 - 8887)
Volume 1 – No. 26
137



mi
i
tfklocidfktfkwk
1
][
(1)
Where wk is the weight of kth term in the document, tfk is the
frequency of the kth term in document, N is the total number of
documents in the collection and idfk is the inverse document
frequency annotated with kth term. loc [tfk] is the weight of the
structural documents field.
3.2 ONTOLOGY DISTANCE MEASURE
The relations between entities are discovered through the measure of
the similarity between the entities of ontology. The ontology based
similarity between sets of concepts helps in retrieving and filtering
information in automatic way.
In our paper, the methods are integrated to find the term relation
information, while these terms are considered to be independent in
the term-based vector space method. In order to find relation
information between terms, first of all, we exploit the background
knowledge which is given through an ontologies source WordNet. A
matcher based on WordNet can be designed by translating the
relations provided by the Wordnet to logical relations according to
the rules of hyponym, hypernym, synonymy, antonymy relations. The
terms s and t are related based on the ontology then the similarity
between the terms is 1 otherwise 0. The simple measure based on
synonymy similarity of Wordnet synsets is given in equation (2).
otherwise
tsif
tsSim
0
0
1),( 


(2)
3.2.1 Resnik Semantic Similarity
Using Resnik semantic similarity [13] each synset of the concept is
associated with a probability of occurrence of an instance of the
concept in a particular concept. The lower probability is assigned to
the more specific concepts. The Resnik semantic similarity considers
the maximum information content and gives more general synset
between the two terms. For Resnik the semantic similarity can be
obtained per the frequency ofappearance in the corpus, and defined
by :
(3)
In (3) IC(Ci) = - log(p(Ci)) is the information content of the concept
Ci (i.e, the entropy of a class Ci). The probability p(Ci) is computed
by dividing the number of instances of Ci by the total number in the
corpus. It provides, however, a systematic way to detect which entity
classes are most similar to each other and, therefore, which entity
classes are the best candidates for establishing integration across the
ontologies. Our similarity measure could be used as a first step
toward a strong integration of ontologies where user input would
provide refinements. This approach is also useful in dynamic
environments, such as the World Wide Web (WWW), where it may
be impractical to force users to subscribe a priori to a shared
ontology.
3.2.2 Minkowski Distance metrics
The similarity between the object can be analysed using Minkowski
distance metrics. In (4) if p=2 then distances are the Euclidean
distance. The distance measure weights give the importance of the
dimension. Based on the importance of the concept the weights are
assigned to the concepts and relations.
(4)
.
3.2.3 Bayes Learning
The query term is matched with the concept in ontology using
Bayes learning method. Based on Bayes rule in (5)
(5)
Where t is the term in query or document and c is the concept in
ontology. The probability of term annotated with the concept is
considered in bayes rule.It provides, however, a systematic way to
detect which entity classes are most similar to each other and,
therefore, which entity classes are the best candidates for establishing
integration across the ontologies. Our similarity measure could be
used as a first step toward a strong integration of ontologies where
user input would provide refinements. This approach is also useful in
dynamic environments, such as the World Wide Web (WWW),
where it may be impractical to force users to subscribe a priori to a
shared ontology.
4. EXPERIMENTAL RESULTS
4.1 Test Collection
The keyword based analysis collects the set of keywords or terms that
occur frequently together and then finds the correlation relationship
among them. The semantic information retrieval KB has been built
and associated to the information document base by using domain
ontology’s that describe the concepts. The query model can employ
to find and manipulate the needful data from the annotated
documents.The performance of the proposed methods are evaluated
using web documents collected from USGS. According to USGS
the topic major kind are environmental contamination, health and
human impacts. The classification can be expressed by hierarchical

©2010 International Journal of Computer Applications (0975 - 8887)
Volume 1 – No. 26
138
structure of the class in the ontology definition. The concepts will be
defined as the document class and some attributes, which describe the
document information. The predefined base ontology described based
on USGS Scientific directory provides the basis for the semantic
indexing of documents with nonembedded annotations. Documents
are annotated with concept instances from the KB by creating
instances of the annotation class. The semantic information retrieval
KB has been built and associated to the information document base
by using domain ontology’s that describe the concepts. Once the
experimental setting has been set up, we have tested the retrieval with
IR functionality in GATE. GATE comes with a full-featured
Information Retrieval (IR) subsystem. In Gate IR the documents can
be retrieved from the corpora not only based on their textual content
but also according to their features or annotations. The current
implementation is based on the most popular open source full-text
search engine – Lucene. The Ontology Annotation Tool (OAT) is a
GATE plug-in available from the Ontology Tools plug-in set, which
enables a user to manually annotate a text with respect to one or more
ontology’s. The required ontology must be selected from a pull-down
list of available ontology’s. OAT also allows users to assign property
values as annotation features to the existing class and instance
annotations.
4.3 Retrieval Performance measure
Our system takes the query and is executed against the knowledge
base and returns the matching documents. A query weight gives the
importance of the concept in the information needed by the user. The
accuracy of the proposed technique has been evaluated against the
result set generated by running the query “ contamination water
pollution”
Several measures such as precision and recall are used to evaluate the
performance of document retrieval. Precision p is defined as the
proportion of retrieved documents that are relevant and is given in (6)
where Ra is the relevant document retrieved and A is the retrieved
document.
A
RaP 
(6)
Recall is defined as the proportion of relevant documents that are
retrieved and is given in (7) where R is relevant document.
R
RaR 
(7)
A is the number of retrieved documents. R is the number of
relevant documents. Ra is the number of retrieved relevant
document. A set of 20 queries was prepared manually for
comparative performance measurement. The set of sample
queries is given in Table 1. Table 2 show the different levels
of performance for different cases the semantic information
retrieval combined with the structural information improves
the document ranking. Fig. 1 shows the performance of
retrieval based on document annotation and without
annotation for the query “minedrainage:Coal”.
Table 1 Average Top-5 Search Ratings for 5 queries
0
0.2
0.4
0.6
0.8
1
0.0 0.2 0.4 0.6 0.8 1.0
pr
e
c
i s
i o
n
Recall
Keyword
Ontology
Figure 2. Comparision between keyword
and ontology distance measure
The Table 2 shows the average weights for calculating
the concepts weight. The Table 3 gives the precision and recall
values. The performance of the two methods is compared
using precision and recall. The F measure is defined using
precision and recall shows the accuracy of the methods by
comparing precision and recall.
Table 2 Average times for calculating
concept weights
Keyword queries Keyword
Rating
Ontology
correlation rating
minedrainage:Coal 2.05 3.31
Contamination
pollution:Water Quality
2.21 3.49
content:Acid Rain 1.25 3.35
environmental pollution 2.70 3.55
toxic 2.34 3.47

©2010 International Journal of Computer Applications (0975 - 8887)
Volume 1 – No. 26
139
Table
3. Best F measure using keyword and ontology
distance measure
Methods Precision Recall F-measure
Keyword 0.737 0.640 0.685
Ontology
distance
0.768 0.727 0.747
Instead of simple keyword index lookup, the semantic search system
processes a semantic query against the KB, which returns the relevant
document. Better precision is achieved by using structured document
annotation weight and the average precision for the top 10 documents
is shown in Table 3. The Table 3 shows that the method can improve
precision by 11% from 0.3761 to 0.4119 in relevant measure. Table
concludes the ontology distance are more accurate compared to
keyword.
5. CONCLUSION
As an extension of the current Web, Semantic Web provides a
structured data and knowledge representation framework for Web
information. Semantic Web provides a structured data and knowledge
representation framework for Web information techniques to generate
metadata that semantically annotating a web page will help
improving lack of semantic information. These similar entity classes
could be then analyzed with user inputs to derive semantic relations,
such as is-a or synonym relations, to create a single, integrated
ontology This paper introduces an annotation scheme that combines
the ontology based similarity measures and concept frequency in the
document Our approach can be seen as a evolution of the keyword
based indices are replaced by ontology based KB and a
semiautomatic document annotation weighting procedure that
improves the retrieval performance.
6. ACKNOWLEDGMENT
The authors express their sincere thanks to the Management
and Principal for their encouragement and support.
7. REFERENCES
1. Ahu SIeg, Bamshad Mobasher and Robin Burke, “Learning
Ontology Based User Profiles: A Semantic Approach to
Personalized Web Search”, IEEE Intelligent Informatics
Bulletin, vol .8, pp.7- 18, 2007.
2. Dik L. Lee, Huei Chauang and Kent Seamons, ” Document
Ranking and the Vector Space Model” , IEEE Software,
pp.66-75, 1997.
3. Guan-yu LI, Sui-ming YU and Sha-sha DAI, “ Ontology based
query system design and implementation”, International
conference on network and parallel computing, pp.1010 -1015,
2007.
4. R.Guha, R.McCool and E.Miller, “Semantic Search”,
International Conference on World Wide Web, pp.700-709,
2003.
5. N. Guarino, C. Masolo, and G. Verete, “OntoSeek: Content-
Based Access to the Web,” IEEE Intelligent Systems, vol. 14,
pp. 70-80,1999.
6. Jerome Euzenat, Pavel Shvaiko, "Ontology Matching",
Springer-Verlag, Berlin Heidelberg(DE),2007,isbn:3-540-
49611-4
7. Jose A.Alonso-Jimenez, Joaquin Borrego-Diaz, Antonia M.
Chavez Gonzalez, Francisco and J. Martin-Mateos,
“Foundational challenges in Automated Semantic Web Data
and Ontology Cleaning”, IEEE Intelligent Systems, pp. 42-52,
2006.
8. J. Jiang and D. Conrath, “Semantic Similarity Based on Corpus
Statistics and Lexical Taxonomy,” Proc. Int’l
Conf.omputational Linguistics (ROCLING X), 1997.
9. J. Lee, M. Kim, and Y. Lee, “Information Retrieval Based on
Conceptual Distance in IS-A Hierarchies,” J.
Documentation, vol. 49, pp. 188-207, 1993.
10. Maedche, A, S.Staab, N.Stojanovic, R.Studer and Y.Sure,
“Semantic portal: The SEAL Approach”, Spinning the
Semantic Web, pp. 317-359, 2003.
11. Mehrnoush Shamsfard, Azadeh Nematzadeh and Sarah Motiee,
“ ORank: An Ontology Based System for Ranking
Documents”, International Journal of Computer Science,
vol .1, pp.225- 231, 2006.
12. Pablo Castells, Mriam Fernandez and David Vallet,” An
Adaption of the Vector Space Model for Ontology based
Information Retrieval”, IEEE Transaction on Knowledge
and Data Engineering, vol. 2, pp.261-22,2007.
13. Philipp Resink.” Semantic similarity in a taxonomy: an
information-based measure and its application to problems of
ambiguity in natural language”, Journal of Artificial Intelligence
Research, Vol 11, pp. 95-130, 1999.
14. A. Smeaton and I. Quigley, “Experiment on Using Semantic
Distance Between Words in Image Caption Retrieval,”
Proc. 19th Int’l Conf. Research and Development in
Information Retrieval SIGIR’96, 1996.
15. Sun Kim and Byoung-Tak Zhang, “Genetic Mining of HTML
Structures for Effective Web Document Retrieval”, Applied
Intelligence, vol.18, pp.243-256, 2003.
16. E. Voorhees, “Using WordNet for Text Retrieval,” WordNet:
An Electronic Lexical Database, C. Fellbaum, ed., Cambridge,
Mass.: The MIT Press, pp. 285-303. 1998.
17. Wang Wei, Payam M.Barjaghi and Andrzej Bargiela, “Semantic
enhanced information search and retrieval”, Sixth International
Conference on Advance Language and Web Information
Technology, pp.218-223,2007.
18. Yufei Li, Yuan Wang, and Xiaotao Huang, “A Relation- Based
Search Engine in Semantic Web”, IEEE Transaction on
Knowledge and Data Engineering, vol.19, pp.273-282, 2007.
Number of
Keywords
2 3 4
Time(msec) 6.31 9.40 12.50