Web personalization using clustering of web usage data

International Journal in Foundations of Computer Science & Technology (IJFCST), Vol.4, No.5, September 2014
WEB PERSONALIZATION USING CLUSTERING OF
WEB USAGE DATA
Bhawesh Kumar Thakur 1, Syed Qamar Abbas2 and Mohd. Rizwan Beg 3
1Department of Computer Science & Engineering, B.I.E.T., U.P.T.U, Lucknow, India
2A.I.M.T.,U.P.T.U Lucknow, India
3Department of Computer Science & Engineering, Integral University, Lucknow, India
ABSTRACT
The exponential growth in the number and the complexity of information resources and services on the Web
has made log data an indispensable resource to characterize the users for Web-based environment. It
creates information of related web data in the form of hierarchy structure through approximation. This
hierarchy structure can be used as the input for a variety of data mining tasks such as clustering,
association rule mining, sequence mining etc.
In this paper, we present an approach for personalizing web user environment dynamically when he
interacting with web by clustering of web usage data using concept hierarchy. The system is inferred from
the web server’s access logs by means of data and web usage mining techniques to extract the information
about users. The extracted knowledge is used for the purpose of offering a personalized view of the
services to users.
KEYWORDS
Web Mining, Web Usage Data, Clustering, Personalization.
1. INTRODUCTION
Data mining gives capabilities to analyzing large amount of data and extract information that are
useful for statistical and business purposes as well. Now a day’s our capability of generating and
collecting data have been increasing exponentially. Contributing factors include, the
computerization of many business, scientific, and government transactions, and advances in data
collection, wide spread use of bar codes, satellite remote sensing data etc. Besides this, heavy
access of the World Wide Web as a knowledge repository provides a large amount of data and
information. This extra ordinary growth in collected data has generated an immediate requirement
for new automated platform that can intelligently help us in converting the huge collection of data
into valuable information and knowledge.
Data mining aims at discovering valuable information that is hidden in conventional databases.
Data mining applied to the web has the potential to be quite beneficial. The emerging field of web
mining aims at finding and extracting relevant information that is hidden in Web-related data.
Etzioni (1996) is believed to be the inventor of ‘Web Mining’ and he described it as: “Web
Mining is the use of data mining techniques to automatically discover and extract information
from World Wide Web documents and services.”
Web mining is the mining of data related to the World Wide Web. It is categorized into three
active research areas according to what part of web data is mined, of which Usage mining, also
known as web-log mining, which studies user access information from logged server data in order
to extract interesting usage patterns [1].
DOI:10.5121/ijfcst.2014.4507 69

Web mining is commonly divided into the following three sub-areas:
1. Web Content Mining: It consists of a number of techniques for searching the Web for
documents whose content meets a certain criterion. It is an Application of data mining techniques
to unstructured or semi structured text.
2. Web Structure Mining: It tries to extract information from the hyperlink structure between
different pages. It uses the hyperlink structure of the web as an (additional) information sources.
3. Web Usage Mining: Web Usage Mining is defined as “the application of data mining
techniques to discover usage patterns from web data”. It is an analysis of user interaction with a
web server.
The commonly used Web mining technologies are user access pattern analysis, Web document
clustering, classification and information filtering. There are some deficiencies such that most of
Web search systems are just in a simple search model, focused on the retrieval efficiency and
ignored the retrieval accuracy due to inherent subjectivity, inaccuracy and uncertainty of user
queries. The soft decision-making is not available. The relevance of the documents is only a part
of attributes and has not very clear boundaries.
Clustering is the process of grouping the data into classes or clusters so that objects within a
cluster have high similarity in comparison to one another, but are very dissimilar to objects in
other clusters. Dissimilarities are assessed based on the attribute values describing the objects.
Clustering has its roots in many areas including data mining, statistics, biology and machine
learning. The entity in data mining could have large number of attributes. Clustering of said high
dimensional data creates major difficulties, more than in supervised learning. In decision trees
irrelevant attributes will not be selected for decision nodes. In clustering high dimensionality
faces a different problem. First, the irrelevant attributes dilutes any indication on clustering
nature. This could also possible with low dimensional space, the probability of occurrence and
number of irrelevant attributes increases with dimension. The second problem is the
dimensionality curse that is a lack of separation of data in a high dimensional space.
Web mining is used to automatically discover and extract information from Web-based data
sources such as documents, log, services, and user profiles.[8] A usage-based Web
personalization system utilizes Web data in order to modify a Web site.[9]
According to Personalization Consortium the main reason of using information technology to
provide personalization are [10]:
• By anticipating needs to cater the customer requirements in a better manner;
• Make the interaction efficient and satisfying for both parties;
• Build a relationship with customer that encourages them to return for consequent visits.
Personalization is a process of collecting and storing information about website users, analyzing
the information relatd to them, and, on the basis of analysis, giving the proper information to each
visitor at the right time.[11]
2. PREPROCESSING OF WEB DATA
In the WWW (World Wide Web) environment, a large amount of traffic can be generated
between clients and web server. The traffic from a client to a server is a URL. The traffic from
server to client is a named HTML file that will be interpreted and displayed on the client screen.
The web usage log probably is not in a format that is usable by mining applications. As with the
data may need to be reformatted and cleaned. Steps that are part of the preprocessing phase
include cleansing, user identification, session identification, path completion, and formatting.
70

• Let P be a set of literals, called pages or clicks, and U be a set of users. A Log is a set of
triplets {(u1,p1,t1)…(un,pn,tn)} where ui U, pi  P, and ti is a time stamp.
Standard log data consist of the following:-Source site, destination site, and time stamp. A
common technique for a server site to divide the log records into sessions.
A session is a set of page references from one source site during one logical period. Historically, a
user logging into a computer, performing work, and then logging off the login would identify a
session and log off represent the logical start and end of the session.
• Data Cleaning
To clean a server log to remove irrelevant items is very important for Web log analysis. The
discovered information are only useful if the data represented in the server log gives a correct
scenario of the user accesses to the Web site. The HTTP protocol maintains an individual
connection for each file requested from the Web server.
• User Identification
Next, unique users must be identified. This is a very complex task because of the existence of
caches, firewalls, and proxy servers. The Web Usage Mining methods that rely on user co-operation
or by the automatic identification of web user are the ways to deal with this problem.
• Session Identification
For logs that cover long duration of time, it is most probable that users will visit the Web site
multiple times. The objectives of session identification are to break the web page visits of each
user into individual sessions. The simplest method for this is through a timeout, where if the time
between page requests crossed a certain threshold, it is supposed that the user is starting a new
session. Many commercial products use 15-30 minutes as a default timeout.
• Path Completion
Another problem in reliably identifying unique user sessions is determining if there are vital
accesses that are not recorded in the web access log. This problem is referred to as path
completion. Methods similar to those used for user identification can be used for path completion.
• Formatting
Once the appropriate pre-processing steps have been applied to the server log, a final preparation
module can be used to properly format the sessions or transactions for the type of data mining to
be accomplished.
Let P = (p1,p2…..pu) is the sequence of Web pages accessed from a certain IP between t1 and tu.
Then , a user session v(t1,tf),tf tu, is defined as: v(t1,tf)=(p1,p2,….pf) : (Δt=tj-ti-1 δ, 1<j f) ∧ (f =
u∨tf+1-tf>δ), where δ is a predefined time threshold [3].
There are many problems associated with the pre-processing activities, and most of these
problems centre around the correct identification of the actual user. User identification is
complicated by the use of proxy servers, client side caching, and corporate firewalls. Cookies can
be used to assist in identifying a single user regardless of machine used to access the web.
3. CLUSTERING HIGH DIMENSIONAL WEB USAGE DATA
Hundreds of properties are there in data mining objects. A terrific complexity is faced in high
dimensional spaces clustering which is much more than predictive learning. For node splitting,
however in decision trees, unnecessary properties would not be selected and those unnecessary
properties would not show any impact on Naïve Bayes. A twofold problem is introduced through
high dimensionality in clustering. Firstly, if there is any desire on clustering tendency,
unnecessary properties occurrence removes it.under whatever definition of similarity,.However, if
71

there are no amount of clusters present then it is a useless process to conduct a search for clusters.
The low dimensional data also follows the same process, Unnecessary properties always rises in
dimension with the propability of its number and occurrence. Secondly, dimensionality curse that
is a loose way of speaking about a lack of data separation in high dimensional space.
Scientifically, nearest neighbour query becomes unbalanced: the distance to the bulk of points is
almost similar to the distance of nearest neighbor. For the dimensions higher than fifteen, this
impact starts showing its brutal effect. Therefore, related to the theory of distance, creation of
clusters is becoming unsure in such conditions.
• Dimensionality Reduction
Most of the clustering algorithms based on spatial datasets indices which helps in rapid search of
the nearest neighbours. As a result, indices provides fine proxies related to dimensionality curse
performance impact. Indices whose dimensions ranges under sixteen work successfully, these
indices are used in clustering algorithms. For a dimension d > 20 their performance degrades to
the level of sequential search (though newer indices achieve significantly higher limits). Hence,
possibly it states that more than sixteen properties data is high dimensional.
Web clustering totally depends on the contents of the site resulting two hundred to one thousand
properties (pages/contents) for reserved Web sites. Dimensions in genomic and biology figures
without any problem goes beyond two thousand to five thousand properties. Finally, thousands of
properties are dealt in data mining and information recovery. Two approaches are used to handle
high dimensionality:
(1) Attributes transformations and
(2) Domain decomposition.
4. CLUSTERING OF WEB USAGE DATA USING CONCEPT HIERARCHY
Clustering of data in a large dimension space is of great interest in many data mining applications.
In this, I present a method for personalizing web user environment using clustering of web usage
data in a high dimensional space based. In this approach, the relationship present in the web usage
data are mapped into a fuzzy proximity relation of user transactions. For this, we cluster the user
transactions using the concept hierarchy of web usage data and similarity relation of user
transactions.
• A user transaction is a sequence of URLS. Let the users’ transaction be:
 = {T1, T2, T3…Tm}
• Let U be the set of Unique URLS appearing in the log data for all transactions
U = {URL1, URL2, URL3…, URLn}
Here, each Ti   is a non-empty subset of U, which is a set of unique URLs appearing in the pre-processed
72
log.
4.1. Concept hierarchy of web usage data
Concept hierarchy plays an important role in knowledge discovery process as it specifies
background and domain knowledge and helps in better mining of data. In proposed approach,
clusters of user transaction are formed from pre-processed log of web usage data, which can be
put in a hierarchy.
• A concept hierarchy is a poset (partially ordered set) (H, < ) where H is a finite set of
concept and < is a partial order on H. [4]

4.2. Clustering user transactions
Clustering of user transactions tend to establish user transactions into groups having similar
characteristics that exhibit similar patterns. In this method, transactions related to user are mapped
into a multi-dimensional space as vectors of URL. Normally clustering technique partition this
space into groups of URL that are nearer to each other based on a distance measure. With
reference to web based transactions, each cluster represents a set of transactions that are similar
based on co-occurrence patterns of URL references [5]. Such knowledge is especially useful in
order to perform market segmentation in E-Commerce application or provide personalized web
content to the users.
• A pair (A, B) is a (formal) concept of (D, T, R) if and only if
A⊆D, B⊆T, A’=B∧B’=A
Where,
A’ = {t∈T | (d,t)∈R for all d∈D }
B’ = {d∈D | (d,t)∈R for all t∈T }[6]
• A proximity relation is a mapping S: D × D → [0, 1] such that for x, y, z  D, the following
rules hold:
S (x, x) = 1 (reflexivity)
S (x, y) = S(y, x) (symmetry)
A fuzzy relation that is reflexive and symmetric is called fuzzy-similarity relation [2].
• - Similarity:-
If S is a proximity relation on D, then given an   [0,1], two elements x, z  D are  -similar
(denoted by x S z) if and only if S (x, y) >=  , definition ( Similarity) are said to -proximate
(denoted by x S+z) if and only if either x S z or there exist a sequence y1,y2,y3,….yn D such
that
x S y1, y1 S y2 , y2 S y3………., yn S z
• Indiscernibility:-
If S: D ×D→ [0, 1] is a proximity relation, then S
+ fulfill the property of reflexivity and symmetry. Based on 
73
+ is an equivalence relation.
From proximity relation concept, S
Similarity condition, the clusters (transactions) will merges, and it appears as it is due to
transitivity.
Hence, S+ is an equivalence relation.
Let the children of a transaction T be denoted as child (T). The similarity of two transactions Ti
and Tj is defined as: - [4]
Sim(Ti, Tj) = Child (Ti)  Child (Tj) / Child(Ti)  Child( Tj )
Let  = {T1, T2, T3…Tm} be the user transactions and U = {URL1, URL2, URL3…URLn}
be the set of unique URLs in all the transactions here Ti  U for I =1 to m. For all Ti, Tj  , I
define a fuzzy relation R on  as
R ( Ti, Tj ) = Sim ( Ti, Tj )
From the concept of Indiscernibility, it can be seen that the similarity of two transactions Ti and
Tj is a number between 0 and 1.When the similarity of two transactions Ti and Tj is 0, then Ti
and Tj are completely dissimilar. On the other hand, if the similarity is 1, then the two
transactions are completely similar.
The measure of similarity gives information about the users’ browsing pattern. Actually, the
browsing of web by any two users may not be exactly similar but may have common interesting
sites. Based on the said definition, a similarity matrix can be calculated and used as the base for

clustering. In the context of clustering user transactions, the huge numbers of dimensions are the
URLs present in the access logs.
We consider the fuzzy relation to decide the amount of similarity between the user transactions.
For all Ti,Tj  , I define a fuzzy relation R on  as:-
R (Ti, Tj) = Sim (Ti, Tj)
Relation R formed by the similarity of user transaction  = {T1, T2, T3…Tm} is a fuzzy proximity
relation and is given by R:
Where ij = Sim (Ti, Tj)
Now, for any given value  >=0, R+ is an equivalence relation. The corresponding partition
represents transaction clusters. These partitions represent the transaction Clusters. The user
specified  value is an important factor based on which the performance of cluster depends. A
domain expert can choose the value of   [0,1] based on her/his experience. Algorithm -1
represents the overall approach used for clustering
74
Table 1. Algorithm-1 for cluster generation based on concept hierarchy using fuzzy similarity
INPUT:
1.  = {T1, T2, T3…Tm} // Set of transactions for unique user
2. U = {URL1, URL2, URL3…URLn} // set of unique URLs appearing in the preprocessed log.
3. Value of  (by domain expert)
OUT PUT:
1. Number of clusters
2. Clusters (Set of Transactions)
Step-1: Find the children (all URLs) in a single transaction T as Child (T)
Step-2: Find the similarity of two transactions. The similarity of two transactions Ti and Tj is
defined as: -
Sim(Ti,Tj)=│Child(Ti)Child(Tj)││Child(Ti)Child(Tj)│
Step-3: Compute similarity matrix for all Ti and Tj , where each value in matrix ij = Sim (Ti, Tj).
Proximity relation is defined as
Sim (Ti,Tj)  [0,1]
Sim (Ti,Ti) = 1 (Reflexivity)
(Ti,Tj) = Sim (Tj,Ti) ( Symmetry)
Value of similarity measure always lies between 0 and 1 (fuzzy similarity)
Step-4 : Compute - Similarity.
For a given value (provided by domain expert) ,here   [0,1], two elements Ti, Tz are  -similar
if Sim (Ti, Tj) >=  , definition (  Similarity) are said to -proximate if there exist a sequence
y1,y2,y3,….yn such that Ti Sim y1, y1 Sim y2 , y2 Sim y3………., yn Sim Tz. (Transitivity)
Step-6: Proximity relation between Ti and Tj fullfil the following properties
1. Reflexivity
2. Symmetry
3. Transitivity
Hence , Sim (Ti ,Tj ) proves the Equivalence Relation. This relation generates the equivalence
classes. For given value  >=0, the corresponding partition (equivalence class) represent transaction
clusters

4.3. Experimental Results & Analysis for Clustering of High Dimensional Data
We assessed the effectiveness of presented new clustering approach by experimenting with real
data sets. The data sets came from MCU Web-site’s server log file. We show similarity matrix
related to one data set and their clusters and partition coefficients in table-I and table-II. The log
data from accesses to the server during a period of 30 days is used for the purpose of generation
of user sessions profiles.
In this model clustering quality is purely dependent on user specific value   [0,1]. Cluster
members vary accordingly variation in the value of   [0,1]. Validity of cluster goodness
produced by this model is highly subjective matter depends upon domain expert.Algorithm-2
gives an approach to check cluster validity.
We evaluate cluster goodness on different value of  on different data sets (six data sets) using
Partition coefficient, which is simple method for evaluating cluster goodness and applicable on
high dimensional data (Web Data). The objective is to seek clustering schemes where most of the
vectors of the dataset exhibit high degree of similarity in one cluster. We note, here, that a fuzzy
clustering is defined by a matrix U =[uij] where uij denotes the degree of similarity of the vector
xi in the j cluster.
Partition Coefficient (Bezdek proposed in Bezdeck et al.(1984) the partition coefficient, which is
defined as[7]
The PC index values range in [1/nc, 1], where nc is the number of clusters. The closer to unity
the index the “crisper” the clustering is. In case that all membership values to a fuzzy partition are
equal, that is, uij=1/nc, the PC obtains its lower value. Thus, the closer the value of PC is to 1/nc,
the fuzzier the clustering is. Furthermore, a value close to 1/nc indicates that there is no clustering
tendency in the considered dataset.
75
Table 2. Cluster validity Algorithm
INPUT:
Nc – Number of clusters
N-Total number of transactions in all clusters
OUTPUT:
PC-Partition coefficient index value
ALGORITHM:
For i =1 to N do
For j = 1 to nc
U = 2
4.4. Snapshot of Web Log data of MCU Server:
2013-12-30 02:47:36
#Fields: date time s-sitename s-computername s-ip cs-method cs-uri-stem cs-uri-query s-port cs-username
c-ip cs-version cs(User-Agent) cs(Cookie) cs(Referer) sc-status sc-substatus sc-win32-
status sc-bytes cs-bytes time-taken
ij

ij denotes the degree of similarity of the vector xi in the jth cluster
end for j
end for i
PC = U/N

2013-12-30 02:47:36 W3SVC8521 NS1 70.84.233.186 GET /robots.txt - 80 - 65.54.188.68
HTTP/1.0 msnbot/1.0+(+http://search.msn.com/msnbot.htm) - - 404 0 2 1814 153 46
2013-12-30 02:47:36 W3SVC8521 NS1 70.84.233.186 GET /forum/Default.asp - 80 -
65.54.188.68 HTTP/1.0 msnbot/1.0+(+http://search.msn.com/msnbot.htm) - - 200 0 0 14115 225
421
2013-12-30 03:01:44 W3SVC8521 NS1 70.84.233.186 GET /Study_Institue_Scheme_2005.htm -
80 - 65.54.188.68 HTTP/1.0 msnbot/1.0+(+http://search.msn.com/msnbot.htm) - - 404 0 2 1814
249 46
2013-12-30 03:40:04 W3SVC8521 NS1 70.84.233.186 GET /forum/www.satpuralawcollege.org
- 80 - 65.54.188.68 HTTP/1.0 msnbot/1.0+(+http://search.msn.com/msnbot.htm) - - 404 0 2 1814
250 93
2013-12-30 11:43:44 W3SVC8521 NS1 70.84.233.186 GET /downloads.htm - 80 - 65.54.188.68
2013-12-30 13:27:27 W3SVC8521 NS1 70.84.233.186 GET /index.htm - 80 - 65.54.188.68
After processing (as per algorithm-1) we obtained following set of transactions (period of 15
minutes) and set of unique URLs for input to algorithm-2.
76
Table 3. User Transactions Detail.
User1 Transaction URLs
65.54.188.68 T1 /robots.txt
/forum/Default.asp
/Study_Institue_Scheme_2013.htm
,, T2 /forum/www.satpuralawcollege.org
,, T3 /robots.txt
/downloads.htm
,, T4 /robots.txt
/index.htm
,, T5 /ResearchProjectPhoto/a2.html
/Media_courses_syllabus_Download_Page.htm
Table 4. Unique URLs Detail.
URLs URL Name
URL1 /robots.txt
URL2 /forum/Default.asp
URL3 /Study_Institue_Scheme_2005.htm
URL4 /forum/www.satpuralawcollege.org
URL5 /downloads.htm
URL6 /index.htm

77
URL7
/ResearchProjectPhoto/a2.html
URL8 /Media_courses_syllabus_Download_Page.htm
Computation of similarity matrix based upon above 5 transactions and 8 unique URLs
Table 4. Similarity Matrix obtained from Test data set-I (text) file, IP Address-65.54.188.68, Date –30-12-
13
T1 T2 T3 T4 T5
T1 1 0 0.25 0.25 0
T2 0 1 0 0 0
T3 0.25 0 1 0.33 0
T4 0.25 0 0.33 1 0
T5 0 0 0 0 1
Table 5. Test Data Set-I for IP Address 65.54.188.68.
IP Address-65.54.188.68, Date –30-12-13
Total Transactions-5, Number of Unique URLS-8
 >= Number of clusters Partition Coefficient (PC)
0.5 5 1.0
0.6 5 1.0
0.7 5 1.0
0.9 5 1.0
5. WEB PERSONALIZATION ARCHITECTURE
The overall architecture of the web personalization model is depicted in figure below. There are
three main components which are emphasized in this figure i.e. Client, Recommendation Engine
and the Server. The client only request for particular URL and in response server provides the
personalized environment i.e. the set of recommended URLs. When a user starts its active session
then the list of URLs gets stored at server’s log file. At the server side, data pre processing is done
using data cleaning, session identification and transaction identification. Structured data output
from the processing of server content data and usage data in a form of transaction file is used to
obtain the transaction cluster using fuzzy logic based similarity relation. This transaction cluster
is then fed to the recommendation engine component. In this component the recommendation set

is generated using its logic. The result of this component provides the set of recommendations
that will be used to personalize the user interface to the server.
78
Figure 1. Architecture for Web Personalization
6. RECOMMENDATION ALGORITHM
For our approach, we use transaction clusters which are obtained from a fuzzy logic based
method. Transaction clustering will result in a set C = {c1, c2, …, ck} of clusters, where each Ci is
a subset of T, i.e., a set of user transactions. Each cluster represents a group of users with
"common" access patterns. Generally transaction clusters are not an effective way of capturing a
combined view of common user profiles. Each transaction cluster may contain thousands of
transactions involving hundreds of URL. We present a method to personalize the user session by
providing a set of URL links dynamically from the highest match score clusters down to lowest
match score cluster.
Let A = {T1,T2,T3,……Tm} be the set of user transactions and
U = {URL1,URL2,URL3,….URLn} be the set of unique URLs in all the transactions here Ti  U
for I =1 to m.For all Ti,Tj  A.
Let C1,C2, …., Cm be the transaction clusters obtained from user transaction A with the help of
fuzzy logic based similarity relation.
STEP-1
For the given input i.e user transactions, accessed URLs and clusters, calculate value of URLi in
Cluster Ck. The value of URLi in Cluster Ck is defined as.
Definition 1: The proportion of number of occurrence of URLi across transactions in the cluster
Ck to the total number of transactions in the cluster is known as the Value of URLi in cluster
Ck.[12]
{T|URLi  Child(T),T Ck}
Val(URLi,Ck) =

79
ΣT {Child(T)| T  Ck
Where Child(T) is the set of children of transaction T.
The measure of value of an URL in a cluster gives the information on the weight of that URL in
that cluster
Let a new transaction is started ,say Tnew. Assign Tnew itself a cluster,that is, Cnew = {Tnew}
The potential usefulness of Ck with respect to Cnew is a factor defining its interestingness can be
estimated by a utility function such as support.
STEP-2
Definition 2: The support of new a cluster Cnew to cluster Ci is defined as
ΣURL(Val(URL,Cnew) – Val(URL, Ci)
Support(Cnew, Ci) =
{ Child(T) | T  Ci  Cnew }
Where Child(T) is the set of children of transaction T. Val(URL,Cnew) is the value of URL in the
cluster Cnew and val(URL,Ci) is the value of URL in the cluster Ci.
Once a new user starts a session, at every step, the objective is to match, the half-way user session
with the suitable clusters and provide correct recommendations to the user.
STEP-3
Definition 3:The match score between Ci and Cnew is defined as
Match(Cnew,Ci) = 1- Support(Cnew,Ci)
The preference set of URLs can be made from the highest match score clusters down to lowest
match score cluster. We can also enforce a threshold  on the matching score to reduce the
dimensional space.
6.1. Experimental Result (User IP Address (65.54.188.68))
We used the access logs from the Web site of the MCRPV (http://www.mcu.ac.in) for our
experiment. The log data from accesses to the server during a period of 30 days was used for the
purpose of generation of user session profiles. The site includes a number of pages related to
university profile and activities held in university campus. After pre processing we prepared
several data sets for experiment purpose.
1. Log Data File: This is pre processed file taken from MCU web log file for the IP address
65.54.188.68.One transaction contains URL accessed by user within 15 minutes.
Table 6. Transactions with their URLs
Transactions URLs
T1 Url1
T1 Url2
T1 Url3
T2 Url4
T3 Url1
T3 Url5
T4 Url1
T4 Url6
T5 Url7
T5 Url8

2. Input Clusters:-These clusters are taken as input using clustering approach having fuzzy
80
logic based proximity relation.
Table 7.Clusters with their transaction
Clusters Transactions
C1 T1
C2 T2
C3 T3
C4 T4
C5 T5
3. New Transaction:- From the web log file I extract a transaction which to be assumed a
new transaction for which we have to provide personalized environment. Initially
transaction contain following unique URLs.
{Url1,Url3}
4. Unique URLs:- These are the unique URLs accessed by user within different
transactions. Each URL i.e. URL1,URL2,URL3 etc. represent a URL entry of web log file
given in table 4.
5. Values of URLs in Clusters:-
Value of URLi in cluster Ci is calculated as follows
Table 8. Value of URLi in Cluster Ci

81
URLs Cluster1 Cluster2 Cluster3 Cluster4 Cluster5 Cluster NEW
Url1 0.333333 0.000000 0.500000 0.500000 0.000000 0.500000
Url2 0.333333 0.000000 0.000000 0.000000 0.000000 0.000000
Url3 0.333333 0.000000 0.000000 0.000000 0.000000 0.500000
Url4 0.000000 1.000000 0.000000 0.000000 0.000000 0.000000
Url5 0.000000 0.000000 0.500000 0.000000 0.000000 0.000000
Url6 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
Url7 0.000000 0.000000 0.000000 0.500000 0.500000 0.000000
Url8 0.000000 0.000000 0.000000 0.000000 0.500000 0.000000
6. Support of new cluster to existing Clusters:-
The support calculation is as follows:.
Table 9. Clusters with their Support
Clusters Support
Cluster 1 0.222222
Cluster 2 0.666667
Cluster 3 0.333333
Cluster 4 0.333333
Cluster 5 0.500000
7. Match score to ClusterNEW to Clusteri :-
The Match score calculation is as follows:
Table 10. Clusters with Match Score
Clusters Match Score
Cluster 1 0.777778
Cluster 2 0.333333
Cluster 3 0.666667

82
Cluster 4 0.666667
Cluster 5 0.500000
Here, highest Match Score corresponds to Cluster1. Hence, user 65.54.188.68 will be provided
Cluster 1 as preference set. i.e. URLs of transactions belongs to cluster1 will be suggested as
preference links for providing the personalized environment.
6.2 Evaluation of Recommendation Effectiveness
In order to evaluate the recommendation effectiveness for given model, We measured the
performance using three different standard measures, namely, precision, coverage, and the F1
measure.[13] These measures are adaptations of the standard measures, precision and recall,
often used in information retrieval.
Table 11. Measures values for Data Sets
Data Set Precision Coverage F1 Measure
Data Set 1
IP Address: 65.54.188.68
0.33 1.0 0.4962
Data Set 2
IP Address: 59.94.43.47
0.5714 1.0 0.7261146
Data Set 3
IP Address:70.84.233.186
0.4705 1.0 0.6394558
Data Set 4
IP Address:61.0.56.2
0.5625 0.9 0.690411
Data Set 5
IP Address:67.180.89.140
0.63636 1.0 0.773
In the context of personalization, precision is used to identify the degree to which accurate
recommendations can be produced by the recommendation engine, while coverage identifies the
ability of the recommendation engine to produce all of the URLs that are likely to be visited by
the user. A low precision in this context will likely result in irritated visitors who are not
interested in the recommendation list, while low coverage will show the inability of the site to
generate relevant recommendations at the moment the user interact with the site. Personalization
systems are often evaluated based on two statistical measures, namely precision and coverage. In
our case of personalization approach, the proposed model is capable of producing
recommendation sets with high coverage and high F1 measure with moderate precision.
7. CONCLUSIONS
In this paper, we proposed a method to construct clusters based on the concept hierarchy and -
similarity. The new model to cluster the web user transactions based on the concept hierarchy of
web usage data and fuzzy proximity relations of user transactions was tested for its effectiveness.
The fuzzy concept ensemble to identify amount of similarities between user transaction clusters.
The user specified  value is an important factor based on which the performance of cluster
depends. A domain expert can choose the value of   [0,1] based on her/his experience. That is,
the approach makes use of domain knowledge for forming clusters and it is very helpful
especially for e-commerce application that uses these user transaction clusters to make
personalized environment for the Web users.

Employing Web usage mining techniques for personalization is directly related to the discovery
of effective profiles that can successfully capture relevant user navigational patterns. In this
paper, we also proposed an approach for automatic discovery of user interest domain. A web site
should permit users to make their choices in natural way. This is possible by giving broad
coverage and understandable choices, so that the user can formulate an easy choice that best
suited to their needs. The proposed model is capable of producing recommendation sets with high
coverage and high F1 measure with moderate precision.
REFERENCES
[1] Lin HuaXu, Hong Liu. Web User Clustering Analysis based on KMeans Algorithm, 2010
83
International Conference on Information, Networking and Automation (ICINA).
[2] Abdolreza Mirzaei and Mohammad Rahmati A Novel Hierarchical-Clustering-Combination Scheme
Based on Fuzzy-Similarity Relations, IEEE Transactions On Fuzzy Systems, VOL. 18, NO. 1,
FEBRUARY 2010.
[3] Dimitrios Pierrakos and Georgios Paliouras, Personalizing Web Directories with the Aid of Web
Usage Data, IEEE Transactions On Knowledge And Data Engineering, Vol. 22, NO. 9, SEPTEMBER
2010.
[4] Quan, Thanh Tho, Hui, Siu Cheug and Cao, Tru Hoang. A Fuzzy FCA-based Approach to
Conceptual Clustering for Automatic Generation of Concept Hierarchy on Uncertainty Data, V.
Sn´a¡sel, R. B¡elohl´avek (Eds.): CLA 2004, pp. 1–12, ISBN 80-248-0597-9.V¡SB – Technical
University of Ostrava, Dept. of Computer Science, 2004.
[5] B.Mobasher, R.Colley and J.Srivastava. Automatic personalization based on web usage mining,
Commun. ACM 43,8 (2000) 142-151.
[6] Yun Zhang, Boqin Feng, Yewei Xue, A New Search Results Clustering Algorithm based on Formal
Concept Analysis, Fifth IEEE International Conference on Fuzzy Systems and Knowledge Discovery,
2008
[7] Halkidi , Maria ,Batistakis, yannis ,et.al., On Clustering Validation Techniques, Journal of Intelligent
Information Systems, 17:2/3, 107– 145, 2001 [2] Gizem, Aksahya & Ayese, Ozcan (2009)
Coomunications & Networks, Network Books, ABC Publishers.
[8] Dragos Arotariteia, Sushmita Mitra. “Web mining: a survey in the fuzzy framework” ELSEVIER,
Fuzzy Sets and Systems 148 (2004)
[9] Magdalini Eirinaki And Michalis Vazirgiannis .”Web Mining for Web Personalization” ACM
Transactions on Internet Technology, Vol. 3, No. 1, February 2003
[10] The Personalization Consortium http://www.personalization.org/personalization.html
[11] Web site personalization http://www 128.ibm.com/developerworks/websphere/library
/techarticles/hipods/personalize.html
[12] B.Mobasher, R.Colley and J.Srivastava. “Automatic personalization based on web usage mining”
Commun. ACM 43,8 (2000)
[13] Bamshad Mobasher, Honghua Dai, Tao Luo and Miki Nakagawa “Discovery and Evaluation of
Aggregate Usage Profiles for Web Personalization” School of Computer Science,
Telecommunications, and Information Systems DePaual University, Chicago, Illinois, USA.
http://ai.stanford.edu/users/ronnyk/WEBKDD-DMKD/Mobasher.pd
Authors
Bhawesh Kuamr Thakur received the M.Tech degree in CSE from the M.C. University,
Bhopal, India. He is currently working toward the PhD degree in computer science at the
Department Of Computer Science & Engineering, Integral University, Lucknow, India.
He is working as a faculty member in the Bansal Institute of Engineering & Technology,
Lucknow, an UPTU, Lucknow affiliated Engineering College. His research interests lie
in the areas of Web mining and Web personalization.

84
Prof. Dr. Syed Qamar Abbas completed his Master of Science (MS) from BITS Pilani.
His PhD was on computer-oriented study on Queueing models. He has more than 20
years of teaching and research experience in the field of Computer Science and
Information Technology. Currently, he is Director of Ambalika Institute of Management
and Technology, Lucknow
Prof. Dr. M. Rizwan Beg is M.Tech & Ph.D in Computer Sc. & Engg. Presently he is
working as Controller of Examination in Integral University Luck now, Uttar Pradesh,
India. He is having more than 16 years of experience which includes around 14 years of
teaching experience. His area of expertise is Software Engg., Requirement Engineering,
Software Quality, and Software Project Management. He has published more than 40
Research papers in International Journals & Conferences. Presently 8 research scholars
are pursuing their Ph.D in his supervision.

Web personalization using clustering of web usage data

Recommended

More Related Content

What's hot (16)

Viewers also liked (19)

Similar to Web personalization using clustering of web usage data (20)

More from ijfcstjournal (20)

Recently uploaded (20)

Web personalization using clustering of web usage data