Ontology oriented concept based clustering

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 02 | Feb-2014, Available @ http://www.ijret.org 600
ONTOLOGY ORIENTED CONCEPT BASED CLUSTERING
Anbarasi .M.S1
, Iswarya.V 2
, Sindhuja.M 3
, Yogabindiya.S 4
1,2,3,4
Department of Information Technology, Pondicherry Engineering College, Puducherry, India
Abstract
Worldwide health centre scientists, physicians and other patients are accessing, analyzing, integrating and storing massive
amounts of digital medical data in different database. The potential for retrieval of information is vast and daunting. The
objective of our approach is to differentiate relevant information from irrelevant through user friendly and efficient search
algorithms. The traditional solution employs keyword based search without the semantic consideration. So the keyword retrieval
may return inaccurate and incomplete results. In order to overcome the problem of information retrieval from this huge amount of
database, there is a need for concept based clustering method in ontology. In the proposed method, WorldNet is integrated in
order to match the synonyms for the identified keywords so as to obtain the accurate information and it presents the concept
based clustering developed using k-means algorithm in accordance with the principles of ontology so that the importance of
words of a cluster can be identified.
Keywords: Ontology, Concept based clustering, K-means algorithm and information retrieval.
---------------------------------------------------------------------***----------------------------------------------------------------
1. INTRODUCTION
The steady and tremendous progress of computer hardware
technology has led to large supplies of powerful and
affordable computers, data collection equipments and
storage media. The technology provides a great boost to
database and makes a huge number of databases available
for information retrieval and data analysis. Due to increasing
and overwhelming amount of information available, there is
a need for tools to automate the information search.
Ontologies help to disambiguate the information and it helps
automatic information processing which provides standard
concepts which relate to specific concepts and terms and
therefore eliminate the irrelevant information. It contains
information about the relationships between the concepts
and it is used to express the semantics of the terms.
Keywords contain the most important information which
emphasizes the entire material. It is considered as the core
operation for processing any text material. Keywords play a
crucial role in extracting the correct information as per user
requirements[3]. Every year thousands of books and papers
are published. It is very hard to manually organise and
analyze it. So there is a good information extraction which
provides the actual contents of a given material. As such
effective keywords are necessary as they express the
importance of the entire material. With the identified
keyword the synonyms are found so that the retrieval of
information will be very accurate. Since different users view
the same thing in different perspective
Clustering is one of the imperative techniques for organising
data in an unsupervised manner. It is particularly used in
handling large data. It is the process of grouping the similar
items together according to some distant measures. It groups
all the data so that the data within the group are more similar
than the ones in other group [6,9]. It makes relevant data
tend to more closely related to each other than the non
relevant data. It is a precise operation used in variety of
information needs
Vector space is one of the most commonly used
representation schemes [7]. The frequency of every single
term is recorded as a vector. The bag of words
representation used for clustering methods is often
unsatisfactory as it ignores relationships such as synonym
between the important terms that do not co-occur literally. In
order to overcome the lack of semantic consideration, the
concept based text clustering method is proposed. In order
identify and extract the concepts, ontologies can be
integrated into the process of clustering as a background
knowledge.
The special requirements of good clustering algorithm are:
1) It should better preserve the relationships between words
like synonyms since there are different words of same
meaning. 2)Associating a meaningful label to each final
clusters is essential.In our proposed system, the user will be
able to find from what type of disease they are suffering
from , by providing symptoms as their inputs and also
theywill be able to find what type treatment they have to
undergo for their disease. Our proposed system not only
considers the keywords but also the concepts (Different
words which shares the same meaning is called synonyms.
Set of these different words that have the same meaning is
called concept)[6]. So our proposed system works well by
matching the synonyms for the given input and makes the
search process effective and for efficient access.
2.PROPOSED METHOD
Proposed system considers not only the keywords but also
the concepts based on the background domain knowledge.
Proposed system is divided into following modules Text
pre-processing, Ontology creation and concept based
clustering.

_______________________________________________________________________________________
Fig-1:OOCBC High Level System Architecture
In the proposed system, Text file and keywords are given as
the input. The keywords are searched in the ontology for its
existence. If it exists, the corresponding parent,
grandparents, children, and its synonyms will be captured.
Based on the keyword, the whole path is retrieved from the
ontology.
2.1 Pre-processing Phase
Patient health report collection is the initial stage for this
phase. The textual information is stored in many kinds of
machine readable form, such as PDF, DOC, HTML, and
XML and so on. After the health reports are collected it is
transformed into TXT format and maintained in the text
files. The system transform the term related into concept
represented one. Mainly, punctuation and special characters
are removed. Finally keywords are extracted in this phase
for further processing[1,2] .
2.2 Medical Ontology Creation
An Ontology is a representation vocabulary, often
specialized to some domain or subject matter. It is a
representation of a set of concepts within a domain and the
relationships between those concepts [4,5].
Its main components are:
 Classes represents concepts which are taken in
broad sense.
 Attributes represents properties of each concept.
 Relations represents the type of association
between concepts of domain.
Here in our system we represent ontology for medical field.
The medical ontology depicts the semantics of the terms
used in the medical field. Medical ontologies are valuable
and effective methods of representing medical knowledge.
In the process of medical diagnosis disease has several
symptoms associated with it. Organising diseases in an
ontology hierarchy is extremely useful as it forms a
pathological classification of diseases for use in medical
systems. So here we create ontology for diseases and
symptoms. We have identified some diseases and we also
have keenly analysed about the symptoms related to each
and every disease. As this is related to medical field, the
query must be perfectly identified and should retrieve a
correct result for it. In any case, a person who constructs the
ontology needs to have some knowledge and prior
experience in ontology construction and some knowledge of
the domain. Usually, domain specialists are discussed to
explain the meaning of domain-specific concepts.
Here there is a need for an effective way to store and
retrieve knowledge related to human diseases. In this
direction ontologies play a crucial role in defining
standardized concepts. Beyond defining standards, medical
ontologies are much more than biomedical vocabularies.
They arrange concepts into ISA and sibling hierarchies,
which effectively relate these concepts in a structural way
that provides valuable inferences upon retrieval. Ontology
hierarchies are valuable methods of knowledge
representation. The most common case in building ontology
is to groundwork the ontology vocabulary on relevant to
medical guidelines. This means that all the relevant data
from the guidelines has to be represented in a methodical
way using a clergy of concepts and relations. Other sources
of medical knowledge include medical articles, other
medical ontologies or terminologies and most importantly,
proficient knowledge .
Fig-2:Ontology Hierarchy with ISA and sibling hierarchies
2.3.Medical Ontology Design
There are two standard approaches in designing ontology.
First one is that tiny particles of the ontology are built first
and then later joined to form the ontology using higher-level
abstract classes. This is the bottom-up approach that is not
implemented often in medical applications. The other way is
to generically design the upper classes and then develop
small parts of the hierarchy, so called top-down approach.
This is used for large medical ontologies as well as
taxonomies. It is advisable to start the process by creating
classes first, then add properties or slots and finally conclude
with individuals [4,5].
Ontology
Creation
Text
File/Keyw
ord
Relevant
information
Concept
based
clustering
Synonyms
Matching

_______________________________________________________________________________________
Ontology is made up of classes, properties or slots,
relationships between classes and individuals. Individuals
are instances or elements of the particular domain. Classes
are composition or groups of individuals. Properties or Slots
are the connection between classes or individuals. An
example of a medical ontology class is «Disease». It is the
super-class of all the Disease types. All the diseases comes
under the class «Disease». The other types of diseases,
example, Diabetes comes under the class «Disease». A class
can be more formal or general (upper class) or more
particular (subclass). For example, A specific class of
«Disease» is «Diabetes». The diabetes types comes under
the class «Diabetes». For example Diabetes type 1 and
Diabetes type 2. All the symptoms of diabetes comes under
the class Diabetes type 1 and type 2 which comes under the
class «Diabetes». In our case the «Disease» class acts as the
most general class. There is no strict and apparent way in
which medical knowledge must be represented. After
creating the class hierarchy, Object properties are created.
The object properties for our medical domain are
hasSymptom and isTreatedBy is specified .
Fig-3:Class Hierarchy
Table-1: Hypertension Disease Class representing
symptoms of hypertension
Disease Class
Name
Object Property Symptoms
Hypertension has_symptom Drowsiness
Hypertension has_symptom Blurred
Vision
Hypertension has_symptom Tinnitus
Hypertension has_symptom Nosebleed
Hypertension has_symptom headache
Hypertension has_symptom Nausea
Hypertension has_symptom Flushing
Hypertension has_symptom Palpitation
Hypertension has_symptom Frequent
urination
Hypertension has_symptom Urgency of
urination
Hypertension has_symptom nocturia
Hypertension has_symptom Dizziness
Hypertension has_symptom Breathing
Difficulty
Hypertension has_symptom Fatigue
Here the object property is created for the disease
hypertension with relevant symptoms. Likewise the class
hierarchy is created for different types of disease and
symptoms relevant to all the diseases.
Fig-4:Class hierarchy for diabetes
Fig-5:Flow Diagram

_______________________________________________________________________________________
Figure 5 represents the flow diagram of our system. First
patient health report is fed into the system and the keywords
are extracted from the report and it is processed so to obtain
the result based on concept based clustering with the
principles of ontology. Here the information is retrieved
based on ontology.
3.CONCEPT BASED CLUSTERING
3.1 Cosine Similarity
The similarity between the different diseases are measured
based on the symptoms. A vector based representative
metric is the cosine measure is used which is defined by the
cosine of the angle between the two vectors:
𝑆𝑖𝑚 𝑉1, 𝑉2 = 𝑉1. 𝑉2 /|𝑉1||𝑉2|--------(1)
where V1 and V2 be term frequency vectors to be compared
for its similarity[7,8]. Initially the cosine similarity values
are taken as centroid for the clusters. Each time centroid
value will change thus the highest similarity value shows
that the symptoms are more related to the cluster.
3.2.K-means Clustering
Concept based clustering is nothing but grouping the
symptoms based upon the concept (Different words which
shares the same meaning is called synonyms. Set of these
different words that have the same meaning is called
concept)[6]. Here clustering process is done based upon the
symptoms with the identified synonyms.
The k-Means algorithm implemented as a simple procedure
that initially selects k random centroids, assigns each
example to the cluster whose centroid is closest, and then
calculates a new centroid for each cluster. Examples are
reassigned to clusters and new centroids are re-calculated
repeatedly until there is no change in clusters[1,9].
4.EXPERIMENTAL RESULTS
The patient reports are collected and it is preprocessed
in order to obtain the symptoms from the report. The
extracted keywords are dynamically mapped with the
online WordNet dictionary. The patient health reports
consists of patient information, data used for analysis, vitals
and their symptoms. In our system, we have to preprocess
the patient report in order to extract the keywords. The
keywords are nothing but the symptoms. We need to extract
the symptoms from the patient report in order to find the
disease and to dynamically map the synonyms for each of
the symptoms extracted from the patient report. Only the
symptoms are needed other details are not needed. Different
users may view the same thing in different perspective. So
user will not be aware of all the words or all the synonyms
with respect to a particular word. Everyone thinks in a very
different manner. For example, user may provide the input
as tiredness, weariness, tire and all the possible meanings
instead of the word fatigue. The result obtained for the user
query is represented in tabular form.
Table-2:Symptoms acquired for the relevant disease
DISEASES SYMPTOMS(represented in
numbers)
Diabetes Type-1 2
Heart Stroke 1
Hypertension 12
The numbers in table1 is nothing but the number of
symptoms relevant to all the diseases. Here the maximum
number of symptoms corresponds to Hypertension where the
symptoms are clustered based on the synonyms. So it will be
identified that hypertension is the relevant disease for the
user’s query and the diagnosis procedure for the relevant
disease will be shown to them. Finally the system is
measured for finding the efficiency of the proposed system.
To evaluate the accuracy of our clustering algorithm, we use
Recall and Precision performance metrics. The value of
precision and recall can be calculated as:
Precision = x/x+y -------------(2)
Recall =x/x+z -------------(3)
where x is the number of total true positives, i.e. the total
number of items clustered together in predefined class and
that are indeed found together by the clustering algorithm. y
is the total number of false positives, i.e. the number of
items not supposed to be found together but are clustered
together and z is the number of total false negatives, i.e. the
number of items which are expected to be found together but
not clustered together by the clustering algorithm. The result
of the experiment based on accuracy is shown in the graph
where it is matched with the highest similarity.
Fig-6:Accuracy of the proposed system
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
200 400 600 800 1000 1200
Recall
Precision

_______________________________________________________________________________________
5.CONCLUSION AND FUTURE WORK
The World Wide Web grows and changes rapidly and many
researchers are stepping into the era of ontology. There is a
highly diverse group of plain text available in free form. The
paper articulates the unique requirements of text clustering
with the support of specific domain ontology. With the use
of ontology, the proposed system is able to categorize the
items on the basis of the concept level. The dimensionality
of the data gets condensed in this proposed concept based
clustering model. It is an efficient method for retrieving the
information even from very large databases. Concept based
clustering exhibits a better performance than the traditional
term based clustering. This system can further be enhanced
for video and image retrieval even from huge repository.
REFERENCES
[1] Hmway Hmway Tar , Thi Thi Soe Nyunt,” Ontology-
Based Concept Weighting for Text Documents”
International Conference on Information Communication
and Management.
[2] HmwayHmwayTar ,ThiThiSoeNyaunt, “ Enhancing
Traditional Text Documents Clustering based on Ontology”,
International Journal of Computer Applications.
[3] Jasmeen Kaur 1, Vishal Gupta, “Effective Approaches
For Extraction Of Keywords”, IJCSI International Journal of
Computer Science Issues, Vol. 7, Issue 6.
[4] Matthew Horridge, ―A Practical Guide To Building
OWL Ontologies Using Protege 4‖ Edition 1.3. The
University of Manchester.
[5] protege.stanford.edu.
[6] Rekha Baghel, Dr. Renu Dhir, “A Frequent Concept
Based Clustering Algorithm”, International Journal of
Computer Applications.
[7] S.Logeswari, Dr.K.Premalatha, “Biomedical Document
Clustering Using Ontology based Concept Weight”,
International Conference on Computer Communication and
Informatics.
[8] Shehata, Fakhri and Mohamed S.Kamel, “An Efficient
Concept Based Mining Model for Enhancing Text
Clustering”, journal of IEEE Transactions on Knowledge
and Data Engineering.
[9] Steinbach M, Karypis G and kumar V, ” A comparison
of document clustering techniques”, KDD Workshop on text
Mining.

Ontology oriented concept based clustering

More Related Content

What's hot (17)

Viewers also liked (18)

Similar to Ontology oriented concept based clustering (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Ontology oriented concept based clustering