Clustering System

Bachelor's thesis (UAS)
Information Technology
Text Mining and Clustering
2013
Prabin Lama
CLUSTERING SYSTEM BASED

ON TEXT MINING USING THE
K-MEANS ALGORITHM
News headlines clustering
BACHELORS THESIS | ABSTRACT

TURKU UNIVERSITY OF APPLIED SCIENCES
Information Technology
December 2013| 45 pages
Instructor: Patric Granholm
Prabin Lama
CLUSTERING SYSTEM BASED ON TEXT MINING

USING THE K-MEANS ALGORITHM
The increasing scope of the web and the large amount of electronic data piling up throughout
the web has provoked the exploration of hidden information from their text content.
News articles published on different news portals throughout the web are the sources of the
information.These can also be very good topics for the research on text mining. Clustering of
similar news headlines and putting them under a single platform with the corresponding links to
the news portal sites can be a very efficient option to the exploration of the same news article
across multiple different news portals, which is, in fact, a tedious and time-consuming task.
This thesis presents the model which analyzes the news headlines across the different news
portals, uses document pre-processing techniques and creates clusters of similar news
headlines.
Data available on the web are structured, semi-structured or unstructured. Webpages are
usually semi-structured because of the presence of html tags. The XML representation of semistructured data facilitates the clustering of similar documents by the use of distance-based
clustering techniques.
News headlines from different news portals are extracted and stored in an XML file. The XML
file is then preprocessed using document preprocessing techniques. Techniques like
tokenization, stop word removal, lemmatization and synonym expansion are used during the
document preprocessing. The selected news headlines are then represented using the vector
space modeling and term-frequency weighting scheme. Finally, the K-means clustering
algorithm is applied to find similarities among the news headlines and create clusters of similar
news headlines. A sample webpage is used to display the clusters of the news headlines with
their corresponding links.
KEYWORDS:
Information extraction, document pre-processing, document clustering, K-means, news article
headlines
CONTENTS
LIST OF ABBREVIATIONS (OR) SYMBOLS
1 INTRODUCTION
2 LITERATURE REVIEW AND BACKGROUND
2.1 Text mining
2.2 Document Pre-processing
2.2.1 Tokenization
2.2.2 Stopword removal
2.2.3 Lemmatization
11
2.2.4 Synonym expansion
12
2.3 Document representation
12
2.4 Information Extraction
13
2.5 Clustering/Organization of documents
13
2.5.1 K-means clustering
14
2.5.2 How does K-means clustering algorithm work ?
14
2.6 Motivation
14
2.6.1 Problem statement
15
2.6.2 Research Question
15
2.6.3 Objectives
16
2.6.4 Scope and Limitations
16
3 REQUIREMENTS ANALYSIS
17
3.1 Data collection
17
3.2 Functional requirement
17
3.3 Non-Functional requirement
18
3.4 Resource requirement
18
4 FEASIBILITY ANALYSIS AND SYSTEM PLANNING
19
4.1 Feasibility analysis
19
4.2 System planning
20
5 RESEARCH METHODOLOGY
21
5.1 Research approach and design strategy
21
5.2 The overall research process
22
6 EXPERIMENTAL SETUP
26
6.1 Corpus
26
6.2 Dataset
26
6.3 Approach
26
7 IMPLEMENTATION
28
7.1 Implementation model
28
7.2 Implementation Phases
29
7.2.1 Flowchart for document pre-processing
30
7.2.2 Flowchart for clustering process
31
8 TESTING AND ANALYSIS
32
8.1 Testing and Analysis with Custom data sample
32
8.1.1Document Input
32
8.1.2 Analysis
41
9 CONCLUSION
42
10 LIMITATIONS AND FUTURE WORK
43
10.1 Limitations
43
10.2 Future work
43
REFERENCES
ERROR! BOOKMARK NOT DEFINED.
APPENDICES
Appendix 1. Sample code for overall clustering and text mining
FIGURES
Figure 1. Use case diagram of the system.
Figure 2. Activity diagram for the system.
Figure 3. Research approach and design strategy of the study.
Figure 4. The overall research process.
Figure 5. Implementation model of the project.
Figure 6. Flowchart for document preprocessing.
Figure 7. Flowchart for clustering component.
Figure 8. Matrix representation of custom data sample.
Figure 9. Graphical representation of custom data.
Figure 10. Graphical representation of Iteration 0 of the K-means algorithm.
Figure 11. Matrix representation of distance between documents and centroids.
Figure 12. Matrix representation of the documents in the cluster.
Figure 13. Graphical representation of Iteration 1 of the K-means algorithm
Figure 14. Representation of distance between documents and new centroids.
17
20
22
23
28
30
31
33
34
35
36
36
37
38
38
39
40
40
TABLES
Table 1. Custom data sample with the term frequency.
32
Table 2. Final result of the custom data after clustering.
41
LIST OF ABBREVIATIONS (OR) SYMBOLS

FCM
Fuzzy C-Means
HTML
Hypertext Markup Language
IE
Information Extraction
LSI
Latent Semantic Indexing
NLP
Natural Language Processing
VSM
Vector Space Model
XML
Extensible Markup Language
1 INTRODUCTION
Along with the development of new smart technologies, the world is going
digital. Because of this, the web is growing day by day and has become a huge
source of the information. Looking up for the precise and relevant information
and extracting it from the web has now become a time-consuming task. There
are many techniques used for the information extraction from the web and text
mining is one of them.
This thesis entitled Clustering System based on Text Mining using the K
means algorithm, is mainly focused on the use of text mining techniques and
the K means algorithm to create the clusters of similar news articles headlines.
The project study is based on text mining with primary focus on data-mining and
information extraction. The news headlines and the links to the different news
portal are fetched via an XML file to the clustering system. The news headlines
within the XML file are then preprocessed using document preprocessing
techniques and finally grouped in the clusters based on their similarities. These
clusters are displayed in a sample webpage with the corresponding links to the
news portal sites.
TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama
2 LITERATURE REVIEW AND BACKGROUND

Existing works on web text mining and clustering are mainly focused on the
different levels like: Web text clustering, Data text mining, Web page information
extraction etc.
Web data clustering researchers Bouras and Tsogkas (2010) proposed an
enhanced model based on the standard k-means algorithm using the external
information extracted from WordNet hypernyms in a twofold manner: enriching
the bag of words used prior to the clustering process and assisting the label
generation procedure following it.
Murali Krishna and Durga Bhavani (2010) proposed the use of a renowned
method, called Apriori algorithm, for mining the frequent item sets and devised
an efficient approach for text clustering based on the frequent item sets.
Maheshwari and Agrawal (2010) proposed centroid-based text clustering for
preprocessed data, which is a supervised approach to classify a text into a set
of predefined classes with relatively low computation and better accuracy.
Qiujun (2010) proposed a new approach to news content extraction using
similarity measure based on edit distance to separate the news content from
noisy information. Jaiswal (2007) performed the comparison of different
clustering methods like K-Means, Vector Space Model (VSM), Latent Semantic
Indexing (LSI), and Fuzzy C-Means (FCM) and selected FCM for web
clustering.
2.1
Text Mining
Text mining, also known as text data mining or knowledge discovery process
from the textual databases, generally, is the process of extracting interesting
and non-trivial patterns or knowledge from unstructured text documents. All the
extracted information is linked together to form new facts or new hypotheses to
be explored further by more conventional means of experimentation.
The classic approach of information retrieval based on keyword search from

WWW makes it cumbersome for the users to look up for the exact and precise
information from the search results. It is up to the user to go through each
document to extract the relevant and necessary information from those search
results. This is an impractical and tedious task .Text mining can be a better
solution for this as it links all the extracted information together, pushes all the
irrelevant information aside, and keeps the relevant ones based on the question
of interest.
2.2
Document Pre-processing
Document pre-processing is the process of introducing a new document to the

information retrieval system in which each document introduced is represented
by a set of index terms. The goal of document pre-processing is to represent
the documents in such a way that their storage in the system and retrieval from
the system are very efficient. Document pre-processing includes the following
stages:
2.2.1 Tokenization
Tokenization in text mining is the process of chopping up a given stream of text

or character sequence into words, phrases, symbols, or other meaningful
elements called tokens which are grouped together as a semantic unit and
used as input for further processing such as parsing or text mining.
Tokenization is a useful process in the fields of both Natural language
processing and data security. It is used as a form of text segmentation in
Natural Language processing and as a unique symbol representation for the
sensitive data in the data security without compromising its security
importance.
Usually, tokenization occurs in a word level but the definition of the word
varies accordingly to the context. So, the series of experimentation based on
following basic consideration is carried for more accurate output:
10
All alphabetic characters in the strings in close proximity are part of one
token; likewise with numbers.
Whitespace characters like space or line break or punctuation characters

separate the tokens.
The resulting list of tokens may or may not contain punctuation and
whitespace
In languages such as English (and most programming languages) where words

are delimited by whitespace, this approach is straightforward. However,
tokenization is more difficult for languages such as Chinese which have no word
boundaries. Simple whitespace-delimited tokenization also presents difficulties
when word collocations such as New York should be treated as one token.
Some ways to address this problem are by developing more complex heuristics,
querying a table of common collocations, or fitting the tokens to a language
model that identifies collocations in a later processing step.
For example:
Input: Friends, Romans and Countrymen
Output: Tokens
Friends
Romans
Countrymen
2.2.2 Stop Word Removal
Sometimes a very common word, which would appear to be of little

significance in helping to select documents matching users need, is
completely excluded from the vocabulary. These words are called stop words
and the technique is called stop word removal.
11
The general strategy for determining a stop list is to sort the terms by
collection frequency and then to make the most frequently used terms, as a
stop list, the members of which are discarded during indexing.
Some of the examples of stop-word are: a, an, the, and, are, as, at, be, for,
from, has, he, in, is, it, its, of, on, that, the, to, was, were, will, with etc.
2.2.3 Lemmatization
Lemmatisation (or lemmatization) in linguistics, is the process of reducing the

inflected forms or sometimes the derived forms of a word to its base form so
that they can be analysed as a single term.
In computational linguistic, lemmatisation is the algorithmic process of getting
the normalized or base form of a word, which is called lemma, using vocabulary
and morphological analysis of a given word. It is a difficult task to implement a
lemmatizer for a new language as the process involves complex tasks such as
full morphological analysis of the word, that is, understanding the context and
determining the role of a word in a sentence (requiring, for example, the
grammatical use of the word).
According to Plisson et al. (2005), lemmatization plays an important role during
the document pre-processing step in many applications of text mining. Beside
its use in the field of natural language processing and linguistics, it is also used
to generate generic keywords for search engines or labels for concept maps.
Lemmatisation and stemming are closely related to each other as the goal of
both processes is to reduce the inflectional forms or derivationally related forms
of a word to its base form. However, stemming is a heuristic process in which
the end of the words or the affixes of the derivational words are chopped off to
receive the base form of the word. Lemmatisation goes through the whole
morphological analysis of the word and uses the vocabulary to return the
dictionary or base form of the word which is called lemma.
12
For instance: If operated on a token saw, stemming may return just s. While
lemmatisation will go through the morphological analysis and return see or saw
depending upon the use of word saw as a verb or noun in the sentence.
2.2.4 Synonym Expansion
According to McCarthy et al. (2007), synonym expansion, also known as lexical
substitution, is the task of replacing a certain word in a given context with
another suitable word similar in meaning.
When a word has multiple meanings, synonym expansion tries to nd the
correct meaning of the word used in a sentence by identifying its synonyms (or
substitutes) in a given context. Given a sentence, for example she was a bright
girl, the task is to nd a synonym that could replace the word bright without
changing the meaning of the sentence. Let us assume that we take bright as
the target word, then the word brilliant could be the suitable substitution for the
selected word, which would both maintain the meaning of the target word and at
the same time t the context of the sentence.
2.3
Document Representation
Document representation is a key process in the document processing and

information retrieval systems. To extract the relevant documents from the large
collection of the documents, it is very important to transform the full text version
of the documents to vector form. A such transformed document describes the
contents of the original documents based on the constituent terms called index
terms. These terms are used in indexing, the relevant ranking of the keywords
for optimized search results, information filtering and information retrieval. The
vector space model, also called vector model, is the popular algebraic model to
represent textual documents as vectors. Using the vector space model,
documents are represented using the term frequency (tf), inverse document
frequency (idf) or tf- idf weighting scheme.
13
2.4
Information Extraction
Information Extraction (IE) is an important process in the field of Natural

Language Processing (NLP) in which factual structured data is obtained from an
unstructured natural language document. Often this involves defining the
general form of the information that we are interested in as one or more
templates, which are then used to guide the further extraction process. IE
systems rely heavily on the data generated by NLP systems. Tasks that IE
systems can perform include:
Term analysis: This identifies one or more words called terms, appearing in the
documents. This can be helpful in extracting information from the large
documents like research papers which contain complex multi word terms.
Named-entity recognition: This identifies the textual information in a document
relating the names of people, places, organizations, products and so on.
Fact extraction: This identifies and extracts complex facts from documents
.Such facts could be relationships between entities or events.
2.5
Clustering / organization of documents
According to Fung, (2001), clustering is the process of grouping contents based

on the fuzzy information, such as words or word phrases in a set of documents.
In simple words, clustering is the process of grouping the set of physical or
abstract objects into classes of similar objects. An object belonging to one
cluster is dissimilar to the object belonging to another cluster.
The importance of document clustering is widely acknowledged by researchers
Qiujun (2010), Jaiswal (2007) and Shah and Elbahesh (2004) for the better
management, smart navigation, efficient filtering and concise summarization of
the large collection of documents such as World Wide Web (WWW).
14
2.5.1 K-means Clustering

There are various methods of clustering. K-means is one of the most efficient
methods for clustering.
From the given set of n data, k different clusters; each cluster characterized with
a unique centroid (mean) is partitioned using the K-means algorithm. The
elements belonging to one cluster are close to the centroid of that particular
cluster and dissimilar to the elements belonging to the other cluster.
2.5.2 How does K-means Clustering Algorithm work?
The letter k in the K-means algorithm refers to the number of groups we want
to assign in the given dataset. If n objects have to be grouped into k clusters,
k clusters centers have to be initialized. Each object is then assigned to its
closest cluster center and the center of the cluster is updated until the state of
no change in each cluster center is reached.
From these centers, we can define a clustering by grouping objects according to
which center each object is assigned to. The detailed algorithm of k-means is
presented in Section 5.2.3.
2.6
Motivation
The web pages contain a huge amount of information in unstructured or semistructured format which can be extracted and transformed to valuable
information as per our requirement. Different techniques are available for this
purpose. Text mining is one of them and is an important step in knowledge
discovery process as already explained above. News articles are one of the
important sources of information which keep people up-to-date with the current
happenings in the world. Sometimes, for the similar headlines of the news
articles, the content may differ across different news portal. The availability of
the similar headlines and their corresponding links to the news portals under a
single roof would be a very efficient and sophisticated option to explore
information on a similar topic.
15
This project provides a better, deeper, and a clearer understanding of the

overall process involved in transforming the headlines extracted as XML form
from the web pages, grouping them into clusters based on the similarity of the
headlines and displaying them under a single platform with the corresponding
links to the original webpages for better exploration of information on a similar
topic across different news portals
2.6.1 Problem Statement
A person who is reading particular news in one news portal may also be
interested in reading similar news coverage on other news portal to find more
about the topic. The problem here is finding the news portal that covers the
similar news. The usual approach is to visit each likely news portal and then
manually look for the similar news in them to find whether the news the person
is looking for is present or not. This is a problematic, time-consuming and
tedious task. This even reduces the users interest in reading that particular
news as well as his enthusiasm to acquire more information on that topic. On
the other hand, people are so much into smart technologies these days that
they always look for the technology that can satisfy their interest without having
them to put in much effort and time.
The solution purposed here is based on the idea of text mining and clustering.
The basic idea is to create clusters of similar news headlines and build news
content warehouse. This clustered information will then be displayed using a
GUI created using asp.net where a reader can find all similar news with the
corresponding links in a single webpage which solves the issues of browsing
each and every news site and looking for the required news in them.
2.6.2 Research Question
The main objective of the project is to answer the question how it is possible to
view and compare similar news which is published in different news portals from
one single platform with the use of text mining?
16
To answer this question, different text mining and clustering techniques are
used to find the similarities between the news headlines and the GUI is created
to display the clusters of similar news headlines with their corresponding links to
the original sites.
2.6.3 Objectives
The objectives of this research work are:
To provide a single platform to place the clusters of similar news

headlines and their corresponding links to the original sites.
To reduce the complexity of visiting each site manually to read the similar
news.
2.6.4 . Scope and Limitation

The scope of this project is the headlines of news articles published in different
news portals. The system is limited in finding the similarities between the news
headlines fetched manually via XML file. Similarity in news content across
different news portals for particular headlines is not covered while making
clusters of the similar headlines.
17
3 REQUIREMENTS ANALYSIS
3.1
Data Collection
In order to carry out the study, the headlines of news articles are collected from
the random online news portals (online) or some random headlines can be
fetched (offline). Those collected data are stored in an XML file for further
document preprocessing. The XML file is then fed to the system by referencing
its location in the local machine.
3.2
Functional requirements
A functional requirement is something the system must do and includes various

functions performed by specific screen, outlines of work-flows performed by the
system and other business or compliance requirements the system has to meet.
3.2.1 Use Case Diagram
Figure 1. Use case diagram of the system.
18
3.3
Non-functional Requirements
The non-functional requirements represent requirements that should work to

assist the project to accomplish its goal. The non-functional requirements for the
current system are:
3.3.1 Interface
The project constructed is console-based. The output is displayed on a console

screen and a sample webpage is created to display the clusters of the news
headlines.
3.3.2 Performance
The developed system must be able to group the given headlines into clusters
based on the K means algorithm.
3.3.3 Scalability
The system must provide as many options like changing headlines in the XML
file and then the changes occur in the clusters as well.
3.4
Resource Requirements
Microsoft Visual Studio for development.
Window XP or greater.
19
4 FEASIBILITY ANALYSIS AND SYSTEM PLANNING

4.1
Feasibility Analysis
The feasibility study is an important issue while developing a system. It deals

with all the specifications and requirements regarding the project and gives the
complete report for project sustainability. The feasibility studies necessary for
the system development are mentioned below:
4.1.1 Economic feasibility
It is the study to determine whether a system is economically acceptable. This
development looks at the financial aspects of the project. It determines whether
the project is economically feasible or not. The system designed is a console
application which needs Visual Studio 2010 and all other hardware
requirements for it so that the start-up investment is not a big issue.
4.1.2 Technical feasibility
The system is developed for general purpose use. Technical feasibility is
concerned with determining how feasible a system is from a technical
perspective. Technical feasibility ensures that the system will be able to work in
the existing infrastructure. In order to run the application, the user only needs to
have Visual Studio 2010 installed and to be able to edit the XML file. These all
requirements can be easily fulfilled.
4.1.3 Operational feasibility
Operation feasibility is concerned with how easy the system is to operate. As it
is a console-based application, it is quite easy to handle with normal Visual
Studio 2010 skills. For the efficient operation of the system, the user needs a
general computer. The GUI is a sample webpage, so it does not require any
special skill to view and click the links. The proposed system is operationally
feasible.
20
4.2
System Planning
4.2.1 Activity Diagram

The activity diagram is the graphical presentation of stepwise computational
and organizational actions (workflows) of a system. The activity diagram for the
system is shown in Figure 2.
Gather News from news portal
Clean the news document
System
Store all news in xml document
Apply document preprocessing
Perform tokenization
Construct document vector
Perform stopword removal

Perform Clustering
Perform lemmatization
Perform Synonym Expansion

Show results
Display portal with clusters of news
Figure 2. Activity diagram for the system.
Display clusters with similar news
21
5 RESEARCH METHODOLOGY
According to Kothari (2004), research comprises defining and redefining
problems, formulating hypothesis or suggested solutions, collecting, organizing
and evaluating data, making deductions and reaching conclusion and further
testing the conclusion whether they fit into formulating hypothesis. Research
methodology may be broadly classified as Exploratory and Conclusive. This
chapter discusses the research methodology and the design strategy
implemented to answer the research question of this project.
5.1
Research Approach and Design Strategy
Since all the document pre-processing techniques and the algorithm planned to
use during this project are existing ones, the research approach is more of
exploratory than conclusive.
The use of existing the K-means algorithm for creating clusters of the news
headlines from the corpus of dataset achieved through various document
preprocessing and text mining techniques and generation of an idea which can
be further explored and carried on to develop a new system related to the web
data mining explains that the research methodology used during this project is
Exploratory.
Data mining is a good research field that uses data either originated from the
researcher for addressing specific problem at hand or secondary data which
have already been collected for other purposes like generating new hypothesis
or new idea based on the existing studies and researches. Secondary data can
be classified as internal and external. The research strategy for this project is
the analysis of secondary data which consists of news articles headlines
gathered from different news portals and stored in an XML file. The research
approach and design strategy of this study is illustrated in Figure 3.
22
Research Design
Exploratory
Quantitative
Text Mining
Secondary Data
External Data
Internet/www
News Portals
Figure 3. Research approach and design strategy of the study.
5.2
The Overall Research Process
In this section, the overall process used for the project study is described. The
process includes step-by-step actions performed for the clustering of the news
headlines from different news portals. The overall research process is illustrated
in Figure 4.
23
Document
Preprocessing
Extracted News
Content
Stop Word
Removal
Tokenization
Document
Clustering
Lemmatization
Document
Representation
(tf vector space
model)
Synonym
Expansion
Preprocessed
Document
Figure 4. The overall research process.

5.2.1 Document pre-processing
When the news content along with their links and news heading is extracted in
an XML document, it is further processed for text mining using document preprocessing techniques. The document pre-processing process includes the
following steps:
5.2.1.1 Tokenization
The detail concept and explanation about Tokenization have been mentioned in
Section 2.2.1.The text used in this project is English. So, instead of using the
complex methods, the tokenization process is accomplished by using space to
split the sequence of characters to token.
5.2.1.2 Stop Word Removal
Sometimes extremely common words which would appear to be of little value in
helping select documents matching a users need are excluded from the
vocabulary entirely. These words are stop words and the process is called stop
word removal. For the purpose of stop word removal, we create a list of stop
24
words such as a, an, the, and prepositions. Hence the tokens contained in the
stop word list are discarded.
5.2.1.3 Lemmatization
The detail concept and explanation about lemmatization have been mentioned
in Section 2.2.3. After the text is tokenized, each inflected term is reduced to its
base form so that it can be analysed as a single term. Lemmatization goes
through the morphological analysis of the word in the text, so it is a preferred
method to stemming.
5.2.1.4 Synonym Expansion
The detail concept and explanation about Synonym Expansion have already
been mentioned in Section 2.2.4. Synonym expansion is carried out by
searching each token in the dictionary and transforming each word to the base
words. The dictionary consists of a list of words and all of their synonyms.
5.2.2 Document Representation
Vector space model is the one of the efficient methods of representing
documents as vectors using the term frequency weighting scheme as
mentioned in Section 2.3. The entire collection of dataset from the XML file is
represented as vectors using the Vector space model.
5.2.3 Clustering Using K-means Algorithm

After the construction of the document vector, the process of clustering is
carried out. The K-means clustering algorithm is used to meet the purpose of
this project.
The basic algorithm of K-means used for the project is as following:
K-means Algorithm
Use:
25
For partitioning where each clusters center is represented by the mean value of
the objects in the cluster.
Input:
k: the number of clusters,
Output:
A set of k clusters.
Method:
Step 1: Choose k numbers of clusters to be determined.
Step 2: Choose Ck centroids randomly as the initial centers of the clusters.
Step 3: Repeat
3.1: Assign each object to their closest cluster center using Euclidean
distance.
3.2: Compute new cluster center by calculating mean points.
Step 4: Until
4.1: No change in cluster center OR
4.2: No object changes its clusters.
5.2.4 Presentation of the results

The results obtained from the entire process, which, in fact are the clusters of
similar news headlines, are presented in the console and also on a sample web
page. The sample web page shows clusters of news headlines with underlying
links to the news portals containing the corresponding news.
26
6 EXPERIMENTAL SETUP
The confidence in the overall system performance is yet to be justified by the
experimental results. This chapter discusses the experimentation carried to
evaluate the clustering system. The requirements for the experimentation are as
following:
6.1
Corpus
The corpus is a large collection of documents. The corpus for this project is the
headlines of the news articles extracted in an XML file along with the
corresponding site names. The headlines are stored as structured text.
6.2
Dataset
The dataset is constructed by retrieving the structured text from the XML file
which consists of the structured news headlines. Then the dataset is fetched to
the document pre-processing model as described in Section 5.2.1.
6.3
Approach
For the experiment, the K-means algorithm is applied on the dataset. The
clusters are created based on the minimum distance between the documents
and the cluster centroids. For each dataset of n clustering examples, n
experiments are run, where each clustering is taken in turn as the single
example test set with the n1 remaining clustering as the training set.
The experiment uses the following logical steps:
1) Read n datasets from the XML file.
2) Apply document preprocessing techniques as discussed in Section 5.2.1 to
every dataset.
3) Create the n-dimensional document vector for the dataset.
4) Pass the document vector to the K-means clustering algorithm.
27
5) Observe the clusters obtained and measure the performance of the model
based the cohesive measures of the clusters.
6) Carry out the experiment for different dataset containing news headlines of
another day.
28
7 IMPLEMENTATION
Implementation is the process of executing a plan or design to achieve some
output. In this thesis, the implementation encompasses the extraction of news
headlines in the XML document, fetching them in the system to go through the
document pre-processing techniques, forwarding the pre-processed documents
to the clustering system and obtaining clusters of similar news headlines as a
final output.
7.1
Implementation model
Figure 5. Implementation model of the project.
29
7.2
Implementation phases
The processes involved during the implementation phases are described in

detail in Section 5.2. The following flowcharts are the diagrammatic
representations of those processes.
30
7.2.1 Flowchart for Document Pre-processing
Start
Input Corpus of
news from xml
document
Tokenize the corpus of

document
Remove stopwords from

corpus
Use lemmatization to find

normalized word
Apply synonym expansion for

each normalized term
Preprocessed
document
Figure 6. Flowchart for document preprocessing.
31
7.2.2 Flowchart for Clustering Process
Figure 7. Flowchart for clustering component.
32
8 TESTING AND ANALYSIS

8.1
Testing and Analysis with custom data sample
K-means is a heuristic method of constructing clusters of the documents. There

is no guaranteed output but the execution time is really fast. The K-means
algorithm constructs the clusters of the documents based on their minimal
distances to the centroid of the cluster. The output depends on optimal
calculation of centroids of the clusters. To explain this, the following custom
sample data is used.
8.1.1 Document Input
The sample data with four documents (D1, D2, D3 and D4) consisting distinct
terms Term 1 indicated as X and Term 2 indicated as Y in the following table is
taken as an input document for the experimental analysis.
Documents
Attribute 1(x) Term 1
Attribute 2(Y) Term 2
D1
D2
D3
D4
Table 1. Custom data sample with the term frequency.
33
The matrix representation of the above data is given below:
Figure 8. Matrix representation of custom data sample.

The graphical representation of the given custom data sample is given below:
34
Figure 9. Graphical representation of custom data.

The input document is now processed for clustering with the use of the Kmeans algorithm. The steps involved in K-means algorithm have already been
explained in Section 5.2.3.
Step 1:Here, K=2 (Number of clusters to be created)
Documents D1 and D2 are assumed as the first cluster centroids.
Let C1 and C2 be the coordinates for D1 and D2 respectively.
So, C1= (1, 1) and C2= (2, 1)
35

Step 2:Distances between each document (D3 and D4) and centroids C1 and C2 are
calculated using the Euclidean distance formula.
For example: Distance from document D3 (4, 3) to first centroid C1 (1, 1) is
Square root of [square of (4-1) + square of (3-1)]
=3.61
and the distance of D3 to C2 = 2.83.
The same method is carried out to calculate distance from D1, D2 and D4 to the
centroids C1 and C2 and the final result is displayed in matrix form.
36
Figure 11. Matrix representation of distance between documents and centroids.

Based on the minimum distance from the centroid, the documents are assigned
to the clusters.
D1 is assigned to cluster 1 as its distance is minimal to C1 compared to C2. D2 is
assigned to cluster 2 as its distance is minimal to C2 compared to C1 and so on
for the other documents.
In the above matrix representation (Figure 12), value 1 in the row indicates that
the document belongs to the corresponding cluster and value 0 represents that
the document is not a member of the cluster.
37
Thus, Cluster 1 has D1 as its member and Cluster2 has D2, D3 and D4 as its
members.
Step 3:In this step, Iteration 1 of the algorithm runs, where the new centroid of each
cluster based on the new membership is calculated.
Cluster 1 has only one member which is D1. So, the centroid remains the same
(C1). Cluster 2 has three members, thus the new centroid is the average
coordinates of the three members (D2, D3 and D4).
C2= ((2+4+5)/3, (1+3+4)/3) = (11/3, 8/3)
38
After the new centroids are computed, the distances of all the documents to
them are computed as in Step 2.

Figure 14 shows the documents and their distances from the new centroids.
Based on the minimal distance to each centroid, the documents are
redistributed to their nearest clusters.

Based on the above document matrix (Figure 15), cluster 1 has two members
D1 and D2 and cluster 2 also has two members D3 and D4.
Step 4:Cluster 1 has two members D1 and D2 and Cluster 2 also has two members D3
and D4. Thus, the new centroid for each cluster is calculated.
39
C1 = Average of coordinates of D1 and D2 belonging to cluster 1.

C2 = Average of coordinates of D3 and D4 belonging to cluster 2.
C1 = ((1+2)/2, (1+1)/2) = (3/2, 1)
C2 = ((4+5)/2, (3+4)/2) = (9/2, 7/2)

All these processes take place in iteration 2.

The distance from each documents to the new centroids are calculated as
described in Step 2.
40

Based on the distances calculated above, the documents are redistributed to
their nearest clusters.

Based on the above document matrix (Figure 18), Cluster 1 has two members
D1 and D2 and Cluster 2 also has two members D3 and D4.
The result obtained from Step 3, represented in Figure 15 and the result
obtained from Step 4, represented in Figure 18, are same. The grouping of
documents from the iteration 1 and the iteration 2 shows that the documents do
not move to new cluster anymore. Thus, the computation of the K- means
algorithm stops. The following result is obtained.
41
Documents
Term1
Term2
Result(Cluster)
D1
D2
D3
D4
Table 2. Final result of the custom data after clustering.

8.1.2 Analysis
From the final result obtained from the test, the conclusion is drawn that the
optimal calculation of centroids is required for the better and more accurate
clustering of the documents. The iteration keeps on going until the stability of
the clusters is obtained.
The real data is the XML documents with a collection of a large number of news
headlines. The clustering process of news headlines involves all the processes
described in Section 8.1.1. The manual calculation of the centroids of all the
documents taken as input for the project and displaying the final results is a
tedious and time-consuming task. So, based on the testing and analysis of the
final output of the custom data sample, the results of the clusters displayed by
the clustering system are analyzed and the conclusion of the project is drawn.
42
9 CONCLUSION
The first goal of the current study was to use text mining techniques on web
news articles headlines. The project study involved the great deal of work on
various areas of information retrieval and text mining and focused on the
various methods for document pre-processing and document clustering.
Text mining and clustering techniques are really powerful. This project study
was completely based on these techniques. The system was created for finding
the similarities among the news articles headlines. Various techniques were
applied for preparing the corpus of a pre-processed document. Lastly, the kmeans clustering algorithm was used for creating the clusters of similar news
articles headlines.
The similar news headlines were grouped into a single cluster. The real world
application of the project study would help people to find the similar news
headlines on different news portal from a single platform. This would not have
been possible without the use of text mining and clustering techniques. In
general, it is not feasible to manually look for similar news in each of the portals
and then compare each of them to find similarities between them.
43
10 LIMITATIONS AND FUTURE WORK

10.1 Limitations
The proposed method only clusters news articles headlines on the basis of their
similarities. The news headlines are not classified into categories such as
national, sports, politics, entertainment, international, etc. We applied K-means
Clustering to construct the clusters containing similar news. The result produced
would have been more accurate if the combined method of clustering such as
hybrid clustering based on partition clustering and hierarchical clustering
methods had been used.
10.2 Future Work
As mentioned in limitations, the clustered news headlines have not been
classified into categories. In future, categorization methods can be added to
render results and user interface more manageable and user friendly. After
successful appreciation of the model, enhancement can be done to cover the
news articles over large domains. Web crawlers and parsers can be developed
to extract information from the whole news portal sites. Text mining and
clustering techniques can be used to create the clusters on the basis of
contents and those contents present on the different news portals can be
displayed under the single platform. That single platform can be any webpage,
web application or any mobile application. In addition, this project study can be
extended to apply the methods proposed for web documents other than articles.
The developed model can be enhanced to use in web content mining.
44
REFERENCES
Ali Shah, N. & M. ElBahesh, E. Topic-Based Clustering of News Articles, University of
Alabama at Birmingham. Retrieved September 23, 2013 from:
http://pdf.aminer.org/000/006/766/topic_based_clustering_of_news_articles.pdf
Bouras, C. & Tsogkas, V. 2010. "W-kmeans: Clustering News Articles Using WordNet".
Retrieved September 5, 2013 from:
http://ru6.cti.gr/ru6/publications/116262780379.pdf
Buscaldi, D., Rosso, P. & Arnal Sanchis, E. 2005. A WordNet-based Query Expansion method
for Geographical Information Retrieval. Retrieved September 2, 2013 from:
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.99.8031&rep=rep1&type=pdf
D. Manning, C., Raghavan, P. & Schtze, H. 2008. Introduction to Information Retrieval,
Cambridge, England: Cambridge University Press. Retrieved September 4, 2013 from:
http://nlp.stanford.edu/IR-book/pdf/irbookonlinereading.pdf
Falinouss, P. 2007. Stock Trend Prediction Using News Articles. Masters Thesis, Lulea
University of Technology, Sweden. Retrieved September 20, 2013 from:
http://epubl.ltu.se/1653-0187/2007/071/LTU-PB-EX-07071-SE.pdf
Fung, G. 2001. A Comprehensive Overview of Basic Clustering Algorithms. Retrieved
September 6, 2013 from:
http://pages.cs.wisc.edu/~gfung/clustering.pdf
Huang, C., Simon, P., Hsieh, S. & Prevot, L. 2007. Rethinking Chinese Word Segmentation:
Tokenization, Character Classification, or Word break Identification. Retrieved September 2,
2013 from:
http://delivery.acm.org/10.1145/1560000/1557791/p69huang.pdf?ip=86.50.66.20&id=1557791&acc=OPEN&key=BF13D071DEA4D3F3B0AA4BA89B
4BCA5B&CFID=389226789&CFTOKEN=39573449&__acm__=1387138322_f651d65355dcc03
5ef1e98e656194624
Jaiswal, P. "Clustering Blog Information" 2007. Master's Thesis Projects, Paper 36, San Jose
State University. Retrieved September 18, 2013 from:
http://scholarworks.sjsu.edu/cgi/viewcontent.cgi?article=1035&context=etd_projects
Kobayashi, M. & Takeda, K. 2000. "Information retrieval on the web". Retrieved September 20,
2013 from:
https://www.ischool.utexas.edu/~i385d/readings/Kobayashi.pdf
Kothari C.R. 2004. Research Methodology: Methods and Techniques. Retrieved October 8,
2013 from:
http://books.google.com/books?id=8c6gkbKiF4C&printsec=frontcover&source=gbs_atb&redir_esc=y#v=onepage&q&f=false
Plisson J., Lavrac N. & Mladenic D. 2004. A Rule based Approach to Word Lemmatization.
Retrieved September 25, 2013 from:
45
http://eprints.pascal-network.org/archive/00000715/01/Pillson-Lematization.pdf
Maheshwari, P. & Agrawal, J. 2010. "Centroid Based Text Clustering". Retrieved October 3,
2013 from:
http://www.ijest.info/docs/IJEST10-02-09-36.pdf
Qiujun, L. 2010. "Extraction of News Content for Text Mining Based on Edit Distance".
Retrieved October 3, 2013 from:
http://www.jofcis.com/publishedpapers/2010_6_11_3761_3777.pdf
Murali Krishna, S. & Durga Bhavani, S. 2010. "An Efficient Approach for Text Clustering Based
on Frequent Itemsets. Retrieved October 2, 2013 from:
http://www.textmining.xpg.com.br/ejsr_42_3_04.pdf
Vectomova, O. & Wang, Y. 2006. "A study of the effect of term proximity on query
expansion" (Abstract). Retrieved October 2, 2013 from:
http://jis.sagepub.com/content/32/4/324.abstract
McCarthy, D. & Navigli, R. 2009. English Lexical Substitution. Retrieved October 4, 2013 from:
http://www.dianamccarthy.co.uk/task10index.html
46
Appendices
Appendix 1.0
Sample code for overall clustering and text mining (NewsClustering.cs).
class NewsClustering
{
//create the list of objects of class headInfo to hold heading detail
public static List<NewsLibrary.NewsInfo> NewsList = new
List<NewsLibrary.NewsInfo>();
public static List<List<NewsLibrary.NewsInfo>> clusterList = new
List<List<NewsLibrary.NewsInfo>>();
static void Main()
{
//pass xml file of news and obtain list of object of news
NewsLibrary.PopulateNews news = new NewsLibrary.PopulateNews();
NewsList = news.populate(@"..\\..\\..\\title.xml");
//pass list of news objects to tokenization and obtain token of headline
NewsLibrary.Tokenization tokenization = new
NewsLibrary.Tokenization();
NewsList = tokenization.Tokenize(NewsList);
//pass list of news objects to stopword and obtain stopword free
headline
NewsLibrary.StopWord stopword = new NewsLibrary.StopWord();
NewsList = stopword.remove(NewsList);
//pass list of news objects to lemmatization and lemmmatize tokens in
headline
NewsLibrary.Lemmatization lematization = new
NewsLibrary.Lemmatization();
NewsList = lematization.Lematize(NewsList);
//past list of news objects to synononym expansion
47
NewsLibrary.SynonymExp Syn = new NewsLibrary.SynonymExp();

NewsList = Syn.FindSynonym(NewsList);
//pass list news objects to KMeansClustering
NewsLibrary.KMean KMean = new NewsLibrary.KMean();
clusterList=KMean.Cluster(NewsList, 15);
//Generate cluster XML file
NewsLibrary.GenerateXML XMLFile = new
NewsLibrary.GenerateXML();
XMLFile.Create(clusterList);
}
Appendix 1

Clustering System

Uploaded by

Copyright:

Available Formats

Clustering System

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Clustering System

Uploaded by

Copyright:

Available Formats

Bachelor's thesis (UAS)

CLUSTERING SYSTEM BASED

BACHELORS THESIS | ABSTRACT

CLUSTERING SYSTEM BASED ON TEXT MINING

2 LITERATURE REVIEW AND BACKGROUND

2.1 Text mining

2.2 Document Pre-processing

2.2.2 Stopword removal

2.2.4 Synonym expansion

2.3 Document representation

2.4 Information Extraction

2.5 Clustering/Organization of documents

2.5.1 K-means clustering

2.5.2 How does K-means clustering algorithm work ?

2.6.1 Problem statement

2.6.2 Research Question

2.6.4 Scope and Limitations

3.1 Data collection

3.2 Functional requirement

3.3 Non-Functional requirement

3.4 Resource requirement

4 FEASIBILITY ANALYSIS AND SYSTEM PLANNING

4.1 Feasibility analysis

4.2 System planning

5.1 Research approach and design strategy

5.2 The overall research process

7.1 Implementation model

7.2 Implementation Phases

7.2.1 Flowchart for document pre-processing

7.2.2 Flowchart for clustering process

8 TESTING AND ANALYSIS

8.1 Testing and Analysis with Custom data sample

10 LIMITATIONS AND FUTURE WORK

10.2 Future work

ERROR! BOOKMARK NOT DEFINED.

Table 2. Final result of the custom data after clustering.

LIST OF ABBREVIATIONS (OR) SYMBOLS

Hypertext Markup Language

Latent Semantic Indexing

Natural Language Processing

Vector Space Model

Extensible Markup Language

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

2 LITERATURE REVIEW AND BACKGROUND

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

The classic approach of information retrieval based on keyword search from

Document pre-processing is the process of introducing a new document to the

Tokenization in text mining is the process of chopping up a given stream of text

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

Whitespace characters like space or line break or punctuation characters

In languages such as English (and most programming languages) where words

2.2.2 Stop Word Removal

Sometimes a very common word, which would appear to be of little

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

Lemmatisation (or lemmatization) in linguistics, is the process of reducing the

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

Document representation is a key process in the document processing and

TURKU UNIVERSITY OF APPLIED SCIENCES THESIS | Prabin Lama

Information Extraction (IE) is an important process in the field of Natural