See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/285837677

E-Learning Platform Usage Analysis

Article in Interdisciplinary Journal of E-Learning and Learning Objects (IJELLO) · January 2011
DOI: 10.28945/1511

CITATIONS READS
39 3,914

4 authors, including:

S. Kontogiannis Ioannis Kazanidis

University of Ioannina Democritus University of Thrace
118 PUBLICATIONS 910 CITATIONS 119 PUBLICATIONS 2,266 CITATIONS

SEE PROFILE SEE PROFILE

Alexandros S. Karakos
Democritus University of Thrace
63 PUBLICATIONS 670 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ioannis Kazanidis on 21 August 2017.

The user has requested enhancement of the downloaded file.

Interdisciplinary Journal of E-Learning and Learning Objects Volume 7, 2011

E-Learning Platform Usage Analysis

Stavros Valsamidis and Sotirios Kontogiannis
Democritus University of Thrace, Greece
svalsam@ee.duth.gr; skontog@ee.duth.gr

Ioannis Kazanidis Alexandros Karakos

TEI of Kavala, Democritus University of Thrace,
Greece Greece
kazanidis@teikav.edu.gr karakos@ee.duth.gr

Abstract
E-learning is technology-based learning, such as computer-based learning, web-based learning,
virtual classroom, and digital collaboration. The usage of web applications can be measured with
the use of indexes and metrics. However, in e-Learning platforms there are no appropriate index-
es and metrics that would facilitate their qualitative and quantitative measurement. The purpose
of this paper is to describe the use of data mining techniques, such as clustering, classification,
and association, in order to analyze the log file of an eLearning platform and deduce useful con-
clusions. Two metrics for course usage measurement and one algorithm for course classification
are used. A case study based on a previous approach was applied to e-Learning data from a Greek
University. The results confirmed the validity of the approach and showed a strong relationship
between the course usage and the corresponding students' grades in the exams.
From a pedagogical point of view this method contributes to improvements in course content and
course usability and the adaptation of courses in accordance with student capabilities. Improve-
ment in course quality gives students the opportunity of asynchronous study of courses with actu-
alized and optimal educational material and, therefore, higher performance in exams.
It should be mentioned that even though the scope of the method is on e-Learning platforms and
educational content, it can be easily adopted to other web applications such as e-government, e-
commerce, e-banking, blogs, etc.
Keywords: E-learning, indexes, metrics, data mining, algorithm.

Introduction
The advances in Information and Com-
Material published as part of this publication, either on-line or munication Technologies (ICT) have
in print, is copyrighted by the Informing Science Institute. introduced e-learning, an alternative
Permission to make digital or paper copy of part or all of these mode of learning, which positively af-
works for personal or classroom use is granted without fee fects the way teaching and learning take
provided that the copies are not made or distributed for profit
place by enabling both educators and
or commercial advantage AND that copies 1) bear this notice
in full and 2) give the full citation on the first page. It is per- students to use their limited time effec-
missible to abstract these works so long as credit is given. To tively (Delacey & Leonard, 2002; Rad-
copy in all other cases or to republish or to post on a server or cliffe, 2002; Starr, 1977). E-learning is
to redistribute to lists requires specific permission and payment technology-based learning, such as
of a fee. Contact Publisher@InformingScience.org to request
redistribution permission. computer-based learning, web-based

Editor: Rowan Wagner

E-Learning Platform Usage Analysis

learning, virtual classroom, and digital collaboration. E-learning describes the ability to electroni-
cally transfer, manage, support, and supervise learning and educational materials (Cone & Robin-
son, 2001; Normark & Cetindamar, 2005). Many authors have discussed the way in which e-
learning can be used for the delivery of training, assessment, and support (Fichter, 2002). E-
learning has made considerable progress since the 1980s, attributable in large measure to technol-
ogical developments.
Nowadays, the variety of different kinds of E-learning systems is very large. There are systems
which support individual learning, collaborative learning, learning content management, learning
activity management, formal learning, informal learning, and workplace learning.
One weakness of the E-learning systems is the lack of exploitation of the acquired information
due to its volume. Most of the time, these systems produce reports with statistical data, which do
not help instructors to draw useful conclusions either about the course or about the students.
Moreover, the existing e-learning platforms do not offer concrete tools for the assessment of user
actions and course educational content. To enable educators to improve their course and learners
to acquire more knowledge, in our experiment two means of course evaluation are used: metrics
and questionnaires.
Server log files store information containing the page requests of each individual user (Ueno,
2002). Data mining techniques have been used to discover the sequential patterns of students' web
usage after the analysis of log files data (Romero & Ventura, 2007). The extraction of sequential
patterns has been proven to be particularly useful and has been applied to many different educa-
tional tasks (Romero, Gutierez, Freire, & Ventura, 2008).
The objectives of this paper are the analysis of the log file of an eLearning system and the deduc-
tion of useful conclusions. Indexes, metrics, and one algorithm for classification, which were
firstly introduced by the authors, are also used (Valsamidis, Kazanidis, Kontogiannis, & Karakos,
2010; Valsamidis, Kontogiannis, Kazanidis, & Karakos, 2010). Finally, data mining techniques
were applied disclosing interesting insights.
The paper initially makes a literature review and follows with the background theory, the pro-
posed methodology, the application of the methodology with the use of a case study relating to
the eLearning, the practical implications of the results, and the conclusions along with future di-
rections.

Literature Review
There are several studies that show the impact of data mining on eLearning. Whilst data mining
methods have been systematically used in a lot of e-commercial applications, their utilization is
still lower in the E-learning systems (Zaiane, 2001). It is important to notice that traditional edu-
cational data sets are normally small (Hamalainen & Vinni, 2006) if we compare them to files
used in other data mining fields such as e-commerce applications that involve thousands of clients
(Srinivasa, 2005). This is due to the typical, relatively small class size although it varies depend-
ing on the type of the course (elementary, primary, adult, higher, tertiary, academic, or/and spe-
cial education); the corresponding transactions are therefore also fewer. The user model is also
different in both systems (Romero & Ventura, 2007).
Very interesting is the iterative methodology to develop and carry out maintenance of web-based
courses, in which a specific data mining step was added (García, Romero, Ventura, & de Castro,
2008). The proposed system finds shares and suggests the most appropriate modifications to im-
prove the effectiveness of a course. The obtained information is used directly by the educator of a
course in order to improve instructional/learning performance. This system recommends the ne-
cessary improvements to increase the interest and the motivation of the students. It is well known

186
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

that motivation is essential for learning: lack of motivation is correlated to learning rate decrease
(Baker, Corbett, & Koedinger, 2004). There are several specialized web usage mining tools that
are used in the e-learning platforms. CourseVis (Mazza & Dimitrova, 2007) is a visualization tool
that tracks web log data from an E-learning system. By transforming this data, it generates graph-
ical representations that keep instructors well-informed about what precisely is happening in dis-
tance-learning classes. GISMO (Mazza & Milani, 2004) is a tool similar to CourseVis, but pro-
vides different information to instructors, such as a student’s details regarding the use of the
course material. Sinergo/ColAT (Avouris, Komis, Fiotakis, Margaritis, & Voyiatzaki, 2005) is a
tool that acts as an interpreter of the students’ activity in an E-learning system. Mostow et al.
(2005) describe a tool which uses log files in order to represent the instructor-student interaction
in hierarchical structure. MATEP (Zorrilla & Álvarez, 2008) is another tool acting at two levels.
Firstly, it provides a mixture of data from different sources suitably processed and integrated.
These data originate from e-learning platform log files, virtual courses, academic and demograph-
ic data. Secondly, MATEP feeds them to a data web house which provides static and dynamic
reports. Analog is another system (Yan, Jacobsen, Garcia-Molina, & Dayal, 1996) which consists
of two main components. The first performs online and the second offline data processing accord-
ing to web server activity. Past user activity is recorded in server log files which are processed to
form clusters of user sessions. In addition, Khribi, Jemni, and Nasraoui (2009) propose an auto-
matic personalization approach. This approach provides online automatic recommendations for
active learners without requiring their explicit feedback. Similar to Analog system, it consists of
two main components: an off-line and an on-line. The off-line component preprocesses the ap-
propriate data in order to model both learner and content. The online component uses the pro-
duced models on-the-fly to recognize the students’ needs and goals, and provides learners with
recommendation lists.
A methodology for the maintenance of web-based courses was also proposed by (Kazanidis, Val-
samidis, & Theodosiou, 2009) which incorporates a specific data mining step. Publications of the
authors relevant to this paper are the automated suggestions and course ranking through a web
mining system (Valsamidis, Kazanidis, et al., 2010) and the proposal of two new metrics, homo-
geneity and enrichment, for web applications assessment, which are also used in this paper (Val-
samidis, Kontogiannis, et al., 2010).

Background Theory
Since the methodology is based on the analysis of the log files of eLearning systems, in the back-
ground theory section the foundations of log files are described, as well as the used indexes and
metrics.

Log Files
Apache web server uses the following configurations for the production of its log files: Common
Log Format (CLF), Extended Log Format (ELF), CooKie Log Format (CKLF) and Forensic Log
Format (FLF) (Pabarskaite & Raudys, 2007). In detail:
1. Common Log Format (CLF): This is the typical format used by several web server applica-
tions. It outputs a format string per HTTP request that describes recorded attributes based on for-
mat symbols that express notations presented in Table 1:
LogFormat "%h %l %u %t \"%r\" %>s %b" common
CustomLog logs/access_log common

187
E-Learning Platform Usage Analysis

Table 1: CLF format symbols and notation

Symbol Notation
%h Client hostname or IP address
%l, %u Client user log name (identified by auth service). If no value is present, a "-" is substituted
%t Date Time of the client system following the format: [dd/MMM/yyyy:hh:mm:ss +-hhmm]
%r Client HTTP request. The HTTP request. The request field contains three pieces of
information. The main piece is the requested resource. The request field also contains the
HTTP method (GET/PUT/POST) and the HTTP protocol version (1.0/1.1).
%s Request status returned by the web server: (200: OK, 3xx : Request redirection error, 4xx:
Client request error, 5xx: Request not found or internal server error that occurred from
request
%b The bytes field is a numeric field containing the number of bytes of data transferred as part
of the HTTP request, not including the HTTP header.

In addition Combined log format (CombLF) is the same as the Common Log Format (CLF), with
the addition of two more fields. The additional fields are:
%{Referrer}i field : The "Referrer" HTTP request header. This gives the site that the
client reports having been referred from. (This should be the page that links to or includes
the current requested object
%{User-agent}i field: The User-Agent HTTP request header. This is the identi-
fying information that the client browser reports about itself.
Combined and Common log formats are considered by the authors as one and shall be referred to
with the CLF notation.
2. Extended Log Format (ELF): A log file in the extended format contains a sequence of lines and
each line may correspond to either a directive or an entry. Entries consist of a sequence of fields
relating to a single HTTP request, similar to CLF fields. Directives record information about the
logging process itself. Lines beginning with the # character contain directives. A typical output of
ELF is the following:
#Software: Microsoft Internet Information Server 4.0
#Version: 1.0
#Date: 1998-11-19 22:48:39
#Fields: date time c-ip cs-username s-ip cs-method cs-uri-stem cs-uri-query sc-status
sc-bytes cs-bytes time-taken cs-version cs(User-Agent) cs(Cookie) cs(Referrer)

1998-11-19 22:48:39 206.175.82.5 - 208.201.133.173 GET /global/images/test.gif - 200

540 324 157 HTTP/1.0 Mozilla/4.0+(compatible;+MSIE+4.01;+Windows+95)
USERID=CustomerA;+IMPID=01234 http://webserver/index.html

The difference between CLF and ELF is that ELF is the CLF derivative of IIS (Internet Informa-
tion Server) logging for Microsoft platforms.
3. CooKie Log Format (CKLF) is part of the Apache web server mod-log-config module and of-
fers additionally to the CLF format the capability of logging cookies content per HTTP request.
The storage of cookies values follows the CLF format under the symbol:
%ic for incoming cookies and %oc for outgoing cookies.
4 .Forensic Log Format (FLF) also follows CLF directives and attributes, logging symbols, and
notation. FLF maintains all CLF symbols and, additionally, keeps two records for each client re-
quest: (a) At the arrival of the request at the web server and (b) after the request has been
processed and a reply has been sent to the client. For the recognition of requests a unique request
ID is assigned per request and a pair of +/- symbols is placed before the ID that signifies the
processing status of the request (logging information has been recorded prior to or after

188
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

processing of the request by the web server). FLF offers the capability of request data setting and
identification. It also offers an estimation of the processing effort of each request and an estima-
tion of the user thinking time since requests coming from the same client session keep the same
ID until session expiration or client IP address or request port change or session (HTTP v1.1
maintains the same client port for client connections in contrast to HTTP v1.0. This means that
user requests cannot be assigned to a specific user based on client request port value but from ses-
sion ID value.). FLF is part of Apache mod_log_forensic module and its forensic capability can
also be used independently by CLF, by using the notation %{forensic-id}n at the apache configu-
ration file. A typical ELF logfile output is the following:
+yQtJf8CoAB4AAFNXBIEAAAAA|GET/manual/de/images/down.gif_HTTP/1.1|Hos
t:localhost|User-Agent:Mozilla/5.0 (X11; U; Linux i686; en-US;
rv%3a1.6) Gecko/20040216 Firefox/0.8
-yQtJf8CoAB4AAFNXBIEAAAAA

The first line logs the forensic ID, the request line and all received headers, separated by pipe
characters (|). The plus character at the beginning indicates that this is the first log line of this re-
quest. The second line just contains a minus character and the same ID. If a request is still pend-
ing or not completed the – sign with the request ID is not recorded at the log file.

Indexes and Metrics

The aforementioned fields in the previous sub-section are not adequate in order to evaluate the
course usage. So, some indexes and metrics are used for the facilitation of the course usage eval-
uation (Table 2). First, the indexes Sessions, Pages, Unique pages, Unique Pages per Coursed per
Session are computed with the use of a Perl program (Kazanidis et al., 2009; Valsamidis, Konto-
giannis, et al., 2010). Then, the metrics Enrichment, Disappointment, Interest and Homogeneity
are calculated.
Table 2: Metrics name and description
Index/Metric name Description of the index/metric
Sessions The total number of sessions per course viewed by users
Pages The total number of pages per course viewed by users
Unique pages The total number of unique pages per course viewed by users
Unique Pages per Course ID The total number of unique pages per course per session
per Session (UPCS) viewed by users
Enrichment The enrichment of courses
Homogeneity The homogeneity of courses
Quality The average of enrichment and homogeneity values

The number of sessions and the number of pages viewed by all users are counted for the calcula-
tion of course activity. Each session reflects when a user logs in to the platform and, after some
activity, logs out from the platform. If there is no activity, there is a timeout of 30 seconds. The
number of pages reflects how many pages were viewed by all users. There are some pages of the
course which were viewed by many users but there were also some other pages not so popular. In
order to refine the situation, we define another index which is called unique pages and measures
the total number of unique pages per course viewed by all users. It counts each page of the course
only once, independently on how many times they were viewed by the users. The Unique Pages

189
E-Learning Platform Usage Analysis

per Course per Session (UPCS) index expresses the unique user visits per course and per session;
it is used for the calculation of the course activity in an objective manner. Because some novice
users may navigate in a course and visit some pages of the course more than once, UPCS elimi-
nates duplicate page visits, since it considers the visits of the same user in a session only once.
Enrichment is a metric which is proposed in order to express the “enrichment” of each course in
terms of educational material. Enrichment is defined as the complement of the ratio of the unique
pages over total number of course web pages as proposed in Valsamidis, Kontogiannis, et al.
(2010):
Enrichment = 1- (Unique Pages/Total Pages) (1)
where Unique Pages<=Total Pages.
Enrichment values are in the range [0, 1). When users follow unique paths in a course this is 0,
while in a course with minimal unique pages this is close to 1. Since it offers a measure of how
many unique pages were viewed by the users, it shows how much information included in each
course is handed over to the end user, inferring that the course contains rich educational material.
Homogeneity metric is another metric, which is defined as the ratio of unique course pages visited
to the number of sessions during which the course was visited.
Homogeneity =Unique pages/Total Sessions (2)
where Total Sessions per course >> Unique course pages.
Homogeneity metric value ranges from [0,1), where 0 means that no user followed a unique path
and 1 that every user followed unique paths. It is a course quality index and characterizes the per-
centage of course information discovered by each user participating in a course.
The aforementioned metrics contribute to the evaluation of course usage. Quality is the average
of the metrics Enrichment and Homogeneity. It is a course quality index and reflects the users
behaviour related to the variance of the educational material.

Methodology
The usage analysis methodology consists of 6 (six) steps: Logging the data, Data pre-processing,
Indexes computation, Metrics calculation, Classifier algorithm, and Data mining. For better com-
prehension of the proposed methodology, this methodology is depicted in Figure 1 and described
below.

Logging the Data

A module that uses FLF format and records attributes before and after web server request
processing was implemented.
This step involves the logging of specific platforms derived from E-learning. In detail, a request
and data recording module, installed at the web server or embedded in the E-learning platform,
records specific e-learning platform fields. More specifically, our installed module at the web
server of the E-learning platform monitors, from different courses, thirteen (13) fields, re-
quest_time_event, remote host, request_uri, remote_logname, remote_user, request_method, re-
quest_time, request_protocol, status, bytes_sent, referer, agent, and user requests, which are rec-
orded with the use of an Apache module, developed in Perl programming language.
The development of such a module has the following two advantages: rapid storage of user in-
formation, since it is executed straight from the server API and not by the E-learning application,
and the produced data are independent of specific formulation used by the E-learning platform.

190
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Start of usage analysis

Logging data

Data pre-processing

Indexes computation

Metrics calculation

Classifier algorithm

Data mining techniques

Grades records

Usage analysis results

Figure 1: Diagram of the steps for usage analysis.

Data Pre-processing
The data of the log file contain noise such as missing values and outliers. These values have to be
pre-processed in order to prepare them for data analysis. Specifically, in this step the recorded
data are filtered. Outlier detection is performed and extremes values are removed. This step is not
performed by the E-learning platform and thus can be embedded into a variety of E-learning sys-
tems. Also, data analysis methods are facilitated for the construction of robust results.
The produced log file is filtered, so it includes only the following three fields: (i) courseID, which
is the identification string of each course; (ii) sessionID, which is the identification string of each
session; (iii) page Uniform Resource Locator (URL), which contains the requests of each page of
the platform that the user visited.

Indexes Computation
The aforementioned fields in the previous sub-section are not adequate in order to evaluate the
course usage. So, the static accessory indexes Sessions, Pages, Unique pages, UPCS are used for
the facilitation of the course usage evaluation (Table 2).

Metrics Calculation
In this step, the metrics Enrichment and Homogeneity are calculated.
The evaluation of courses usage is based on two metrics: the metric Quality and the metric Final.
The metric Quality is the mean of Enrichment and Homogeneity, while the metric Final is the
product of Quality with UPCS. These two metrics allow us to classify and group the courses de-
pending on their usage in the electronic platform.
The evaluation of the courses is examined initially with UPCS, which is a quantitative index. It is
quantitative metric because it measures precisely the number of instances. The courses with a
high value of UPCS are popular enough among the students.

191
E-Learning Platform Usage Analysis

Since some courses have the same value of UPCS or values very close to each other, we wanted
to clarify the situation and add one absolutely qualitative metric, that is named Quality, which
combines appropriately the metrics Enrichment and Homogeneity with equal weights. The total
result derives from the product of Quality with UPCS.

Classifier Algorithm
Online courses incorporate specific characteristics according to their educational material. Au-
thors should be able to assess and improve their courses so as to provide a better learning expe-
rience for learners. In this section an algorithm which classifies online courses according to spe-
cific metrics is presented, The Classifier algorithm may be used with online tools in order to au-
tomatically provide feedback to authors about their courses along with useful suggestions for im-
provement. Some crucial factors of an online course are the quantity of each course educational
material, content update, and, of course, acceptance and usefulness. These are the factors used
with Clacifier Algorithm in order to classify online courses into specific groups with identical
characteristics. In order to succeed in this it makes use of specific metrics which are proposed
earlier in this paper.
More specifically, Classifier algorithm classifies courses through three distinct stages. The first
stage of the algorithm aims to identify how rich or poor the educational content is through the use
of the Enrichment metric. A high Enrichment value indicates a course with rich educational con-
tent while correspondingly a low Enrichment value points to a course with poor content that
needs additional material to be added. In order to classify courses at this stage they are sorted in
descending order according to their Enrichment value and are characterized either as courses with
high or low Enrichment. Those courses whose Enrichment is higher than the average Enrichment
of all courses are characterized as high Enrichment courses while the others as low Enrichment
courses. This stage leads to two clusters of courses: those with rich and those with poor educa-
tional material respectively.
The second stage of the algorithm tries to spot how often course information is added or updated
by educators. At this stage the algorithm further classifies the previous set of courses using the
Homogeneity value. The higher the Homogeneity value the more frequently the course updates or
the more dynamic the course content, depending on Enrichment value. The lower the Homo-
geneity value then the LMS is more static in content or of poor content updates. The classification
of the courses at this stage depends on both the average Enrichment value and the average Homo-
geneity value of the high and low Enrichment clusters accordingly. Therefore, this stage of the
algorithm results in four clusters.
The third stage takes into consideration UPCS value in order to check whether users find each
course useful and as a consequence visit its pages. Therefore, the previous clusters were further
split into high and low UPCS courses according to UPCS average value.
The outcome of the algorithm execution is eight clusters of courses. Each cluster is constituted by
courses with similar characteristics. The Classifier algorithm provides a description of each clus-
ter courses (Table 3) so that educators may be informed about the characteristics of their courses
and make the appropriate improvements.

192
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Table 3: Course classification according classifier algorithm

Cluster
Enrichment Homogeneity UPCS Course Classification
ID
I Frequently Frequent course content updates visited by
High High
Updated users
II Frequently Frequent course content updates not visited by
High Low
Updated users
III High Static Content High Static content visited frequently by users
IV High Static Content Low Static content visited occasionally by users
V Frequently
Low High Abandoned course open for view
Updated
VI Frequently
Low Low Garbage course that needs further justification
Updated
VII Poor content that still contains useful informa-
Low Static Content High
tion for students
VIII Low Static Content Low Abandoned course

Data Mining Techniques

Data mining techniques have been applied to E-learning systems data by many researchers. Apart
from the analytical review by Romero and Ventura (2007), there is some more domain specific.
Castro, Vellido, Nebot, and Mugica (2007), among others, deal with the assessment of the stu-
dents’ learning performance, provide course adaptation and learning recommendations based on
the students’ learning behaviour, deal with the evaluation of learning material and educational
web-based courses, provide feedback to both teachers and students of e-learning courses, and
detect a typical student’s learning behaviour. Another similar study described the detection of
typical behaviour in the grouping structure of the users of a virtual campus (Castro, Vellido, Ne-
bot, & Minguillon, 2005). A survey by Koutri, Avouris, and Daskalaki (2005) provides an over-
view of the state of the art in research of web usage mining, while discussing the most relevant
criteria for deciding on the suitability of these techniques for building an adaptive web site. Two
relevant studies, one (Kotsiantis, Pierrakeas, & Pintelas, 2004) predicts the students’ performance
as well as to assess the relevance of the attributes involved and the other (Minaei-Bidgoli, Kashy,
Kortemeyer, & Punch, 2003) predicts student performance by applying data mining methods in
an educational web-based system.
Students are assessed in the final exams of the courses, and they are assigned a grade according to
their performance on the course. Having measured the students' activity of the E-learning system
according to the indexes and metrics, it is possible to investigate whether there is a relationship
between student activity in the platform of the E-learning system and the grades of the students in
the final exams.
In this sub-section we try to discover the relationship between the attribute Grade and the other
attributes shown in Table 2.
There are various classification methods. Kotsiantis, Pierrakeas, and Pintelas (2003) compared six
classification methods (Naive Bayes, decision tree, feed-forward neural network, support vector
machine, 3-nearest neighbour, and logistic regression) to predict drop-outs in the middle of a
course. In the classification step, the algorithm 1R (Witten & Frank, 2005) may be applied. It uses
the minimum-error attribute for prediction, discretizing numeric attributes (Holte, 1993). The
attribute Grade has to be used as class since it describes the education outcome. In this step the
attribute/s which best describe the classification will be discovered.

193
E-Learning Platform Usage Analysis

The clustering step contains course clustering, based on the Grade attribute. Clustering of user
visits is performed with the use of k-means algorithm (MacQueen, 1967), an efficient partitioning
algorithm that decomposes the data set into a set of k disjoint clusters. It is a repetitive algorithm
in which the items are moved among the various clusters until they reach the desired set of clus-
ters. With this algorithm both a large degree of similarity for the items in the same cluster and a
noticeable difference in items which belong to different clusters are achieved. The Manhattan
distance will be used instead of the default Euclidean distance, so that the centroids will be com-
puted as the component-wise median rather than mean. The number of clusters is proposed to be
2, since our goal is to separate the courses into high activity and low activity ones.
Association rule mining is one of the most well studied data mining tasks. It discovers relation-
ships among attributes in databases, producing if-then statements concerning attribute-values
(Agarwal, Imielinski, & Swami, 1993). An association rule X ⇒ Y expresses a close correlation
among items in a database, in which transactions in the database where X occurs, there is a high
probability of having Y as well. In an association rule X and Y are called respectively the antece-
dent and consequent of the rule. The strength of such a rule is measured by values of its support
and confidence. The confidence of the rule is the percentage of transactions with antecedent X in
the database that also contain the consequent Y. The support of the rule is the percentage of
transactions in the database that contain both the antecedent X and the consequent Y in all trans-
actions in the database.
The Weka system has several association rule-discovering algorithms available. The Apriori algo-
rithm (Agarwal, Mannila, Srikant, Toivonen, & Verkamo, 1996) will be used for finding associa-
tion rules over discretized LMS data. Apriori (Agrawal & Srikant, 1994) is the best-known algo-
rithm to mine association rules. It uses a breadth-first search strategy to count the support of item
sets and uses a candidate generation function that exploits the downward closure property of sup-
port. It iteratively reduces the minimum support until it finds the required number of rules with
the given minimum confidence.
There are different techniques of categorization for association rule mining. Most of the subjec-
tive approaches involve user participation in order to express, in accordance with his/her previous
knowledge, which rules are of interest. One technique proposes the division of the discovered
rules into three categories (Minaei-Bidgoli, Tan, & Punch, 2004). (1) Expected and previously
known: This type of rule confirms user beliefs, and can be used to validate our approach. Though
perhaps already known, many of these rules are still useful for the user as a form of empirical ve-
rification of expectations. For education, this approach provides opportunity for rigorous justifi-
cation of many long held beliefs. (2) Unexpected: This type of rule contradicts user beliefs. This
group of unanticipated correlations can supply interesting rules, yet their interestingness and
possible action ability still requires further investigation. (3) Unknown: This type of rule does not
clearly belong to any category and should be categorized by domain specific experts.

Case Study
In this section, the results after the application of usage analysis methodology will be described.
The dataset was collected from a real E-learning environment used at the Kavala Institute of
Technology that operates the Open eClass e-learning platform (GUNet, 2011). The data are from
the spring semester of 2009 from the Department of Information Management and involve 1199
students and 39 different courses. The data are in ASCII form and are obtained from the Apache
server log file. A view of the collected data is shown in Table 4.

194
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Table 4: eClass data from log file

remote_host request_uri re- re- re- request_time re- status bytes refer Agent
mote_l mote_u quest_ quest_p _sent rer
og- ser method rotocol
name
66.249.72.212 /component/search/ - - GET [03/Mar/2009: HTTP/1. 200 9805 - Mozil-
smb.conf.html 18:57:00 1 la/5.0(compatible
+0200] ; Googlebot/2.1;
+http://ww...
66.249.72.212 /newsfeeds.html - - GET [03/Mar/2009: HTTP/1. 200 7967 - Mozil-
18:58:01 1 la/5.0(compatible
+0200] ; Googlebot/2.1;
+http://ww...

As described in the second step of the methodology, the produced log file is filtered and pre-
processed in order to include the following fields: courseID, sessionID and page Uniform Re-
source Locator (URL).
In the third and the fourth steps, the indexes are computed and the metrics are calculated.
The results for the 39 courses are presented in Table 5. The data are ranked in descending order in
the column Final Score.
Table 5: E-Learning data and grade for 39 Courses
Course ID Sessions Pages Unique pages UPCS Enrichment Homogeneity Quality Final score Grade
IMD105 91 297 11 216 0.963 0.121 0.542 117.055 6.78
IMD35 87 338 8 179 0.976 0.092 0.534 95.612 6.14
IMD132 152 230 7 184 0.970 0.046 0.508 93.437 5.75
IMD36 72 217 7 134 0.968 0.097 0.532 71.353 6.16
IMD125 93 164 6 134 0.963 0.065 0.514 68.871 6.13
IMD129 75 209 6 131 0.971 0.080 0.526 68.860 6.21
IMD41 98 185 8 129 0.957 0.082 0.519 66.976 5.87
IMD66 56 144 9 107 0.938 0.161 0.549 58.754 5.96
IMD17 53 206 11 89 0.947 0.208 0.577 51.360 6.83
IMD111 33 142 9 79 0.937 0.273 0.605 47.769 6.96
IMD8 45 135 8 82 0.941 0.178 0.559 45.859 5.34
IMD11 51 108 6 80 0.944 0.118 0.531 42.484 6.11
IMD44 48 82 10 75 0.878 0.208 0.543 40.739 5.93
IMD26 50 90 8 71 0.911 0.160 0.536 38.024 5.99
IMD61 32 74 9 64 0.878 0.281 0.580 37.108 5.78
IMD98 32 113 9 61 0.920 0.281 0.601 36.649 6.96
IMD62 23 94 11 52 0.883 0.478 0.681 35.392 7.12
IMD14 45 122 7 59 0.943 0.156 0.549 32.396 5.82
IMD133 25 80 7 54 0.913 0.280 0.596 32.198 6.75
IMD34 38 113 6 56 0.947 0.158 0.552 30.934 6.21
IMD115 18 73 9 42 0.877 0.500 0.688 28.911 7.15
IMD9 26 105 12 42 0.886 0.462 0.674 28.292 7.10
IMD112 30 62 6 46 0.903 0.200 0.552 25.374 6.18
IMD122 33 71 7 45 0.901 0.212 0.557 25.054 6.23
IMD120 38 80 3 46 0.963 0.079 0.521 23.953 6.07
IMD64 22 47 7 39 0.851 0.318 0.585 22.800 6.23

195
E-Learning Platform Usage Analysis

Course ID Sessions Pages Unique pages UPCS Enrichment Homogeneity Quality Final score Grade
IMD80 14 38 7 34 0.816 0.500 0.658 22.368 6.89
IMD60 22 43 5 40 0.884 0.227 0.555 22.220 6.02
IMD50 22 46 7 38 0.848 0.318 0.583 22.154 6.11
IMD10 17 61 8 28 0.869 0.471 0.670 18.752 7.03
IMD114 28 42 4 34 0.905 0.143 0.524 17.810 5.79
IMD21 11 25 8 24 0.680 0.727 0.704 16.887 6.95
IMD23 30 38 4 32 0.895 0.133 0.514 16.449 6.51
IMD96 20 31 5 30 0.839 0.250 0.544 16.331 5.83
IMD130 12 30 5 22 0.833 0.417 0.625 13.750 6.71
IMD134 25 27 4 27 0.852 0.160 0.506 13.660 6.42
IMD15 11 24 7 20 0.708 0.636 0.672 13.447 6.79
IMD49 14 23 5 21 0.783 0.357 0.570 11.967 6.03
IMD67 18 23 4 22 0.826 0.222 0.524 11.531 5.71

Application of Classification Algorithm

The Classification algorithm classifies the courses. The 39 courses were initially ranked accord-
ing to the Enrichment metric. The algorithm was tested by picking the best and worst 12 LMS
courses from a list of 39 courses, which are shown in Table 6. That is, best and worst cases from
students’ usage point of view.
Table 6. Processed data for 12 Courses with Average Enrichment value of 0.898
Course ID Sessions Pages Unique pages UPCS Homogeneity Enrichment
IMD132 152 230 5 184 0.033 0.978
IMD35 87 338 9 179 0.103 0.973
IMD125 93 164 6 134 0.065 0.963
IMD129 75 209 8 131 0.107 0.962
IMD105 91 297 12 216 0.132 0.960
IMD41 98 185 8 129 0.082 0.957
IMD36 72 217 10 134 0.139 0.954
IMD17 53 206 21 89 0.396 0.898
IMD66 56 144 16 107 0.286 0.889
IMD8 45 135 18 82 0.400 0.867
IMD122 33 71 21 45 0.636 0.704
IMD112 30 62 20 46 0.667 0.677

Based on the previous order by Enrichment Table 5 of 12 LMS courses, the Classifier algorithm
was applied by using an average Enrichment value of 0.898 and average homogeneity value for
the high enrichment cluster of 0.09 and for the low enrichment cluster of 0.45. The classification
of the algorithm produced four clusters, which are shown in Table 7.

196
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Table 7: Clustering of the 12 courses based on the Classifier algorithm

Enrich- Homo-
Course Unique
ment geneity Sessions Pages UPCS Homogeneity Enrichment
ID pages
Class Clusters

Dynamic IMD105 91 297 12 216 0.132 0.960

Content or IMD35 87 338 9 179 0.103 0.973
Frequently
Updated, IMD36 72 217 10 134 0.139 0.954
Cluster I IMD129 75 209 8 131 0.107 0.962
Static Con- IMD132 152 230 5 184 0.033 0.978
tent with
IMD125 93 164 6 134 0.065 0.963
frequent
updates,
Cluster II IMD41 98 185 8 129 0.082 0.957
High

Dynamic IMD112 30 62 20 46 0.667 0.677

Content
with less
updates, IMD122 33 71 21 45 0.636 0.704
Cluster III
IMD66 56 144 16 107 0.286 0.889
Static Con-
tent, Clus- IMD17 53 206 21 89 0.396 0.898
ter IV
Low

IMD8 45 135 18 82 0.400 0.867

As shown in Table 7, for each one of the four classes the LMS courses are ordered based on the
UPCS metric value. So courses IMD105 and IMD35 are the representatives of Cluster I which
means courses with rich educational content which is frequently updated and high UPCS values.
Respectively IMD36 and IMD129 belong to Cluster II and have identical characteristics with first
cluster courses but low UPCS value. In Table 8, these courses and the classifier algorithm evalua-
tion feedback for each one of these courses are presented.
Table 8: Clustering of the 12 courses based on the classifier algorithm
Cluster Course CCA Evaluation
ID ID
I IMD105, High Activity LMS with updates followed by users
IMD35
II IMD36, High Activity LMS with frequent educator updates that are not followed by users
IMD129
III IMD132 High Activity LMS with Static content, frequently updated and followed by users
IV IMD125, High Activity LMS with Static content, frequently updated but poorly followed
IMD41 by users
V - Abandoned course of dynamic content, open for view
VI IMD112, Garbage course or Forum with updates- Need for further evaluation
IMD122
VII IMD66 Course of poor static content that still contains information followed by users (or
forced to follow)
VIII IMD17, Abandoned course of poor static content occasionally followed by curious users
IMD8

197
E-Learning Platform Usage Analysis

Data Mining Techniques

The application of the data mining techniques was achieved with the use of the open source data
mining tool Weka (Witten & Frank, 2005). Weka is open source software that provides a collec-
tion of machine learning and data mining algorithms for data pre-processing, classification, re-
gression, clustering, association rules and visualization. The attributes of Table 5 were inserted in
.cvs format into Weka. The attributes Course ID, Homogeneity, and Enrichment were removed
before the application of classification, clustering, and association. Course_ID is different for
each instance, and Homogeneity and Enrichment are dependent on the other attributes by formu-
las. All the remaining attributes were discretized.

Classification
In the classification step, the algorithm 1R is applied. The attribute Grade is used as class. The
results (Figure 2) show that the best attribute which describes the classification is Quality. This
means that Quality is more closely related to Grade than the other variables and therefore courses
with higher usage help students to achieve better grades and improve their educational perfor-
mance.

Figure 2: Classification using Grade as class.

Clustering
The clustering step contains course clustering, based on the Grade attribute with the use of the
SimpleKmeans algorithm (Kaufmann & Rousseeuw, 1990; MacQueen, 1967). The number of
clusters is defined as 2 and the used distance is Manhattan instead of the default Euclidean dis-
tance. The produced results (Figure 3) show that 13 (33%) of the courses had high activity and 26
(67%) of the courses had low activity. Since previously presented results show that quality is
closely related to the students’ grades, authors should improve their courses in order to increase
their quality and as a consequence the course usage by the students.

198
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Figure 3: Clustering using Grade as class.

Association rule mining

The Apriori algorithm (Agarwal et al., 1996) was used to find association rules over the discre-
tized E-Learning data Table 3 of the 39 courses, executing this algorithm with a minimum sup-
port of 0.1 and a minimum confidence of 0.9 as parameters.
Because of the obvious dependencies of the attributes Sessions, Pages, and Unique Pages with the
attributes Enrichment and Homogeneity, the latter group of attributes was removed from the data
table. Weka shows a list of 15 rules (Figure 4) with the support of the antecedent and the conse-
quent (total number of items) at 0.1 minimum, and the confidence of the rule at 0.9 minimum
(percentage of items in a 0 to 1 scale).

Figure 4: Weka executing Apriori algorithm based on confidence metric.

The application of the Apriori algorithm for association provided useful insights into the LMS
usage. Figure 4 shows how a large number of association rules can be discovered. There are also
some uninteresting and random rules, like rule 9, or redundant rules (rules with a generalization

199
E-Learning Platform Usage Analysis

of relationships of other rules, like rule 5 with rules 1 and 2, and rule 13 with rules 7 and 8).
There are some similar rules, rules with the same element in antecedent and consequent but inter-
changed, such as rules 12 and 13, and 3, 4, and 5. There is also a similar couple of rules, such as
rules 12 and 13. But there are also rules that show relevant information, like those that show ex-
pected or conforming relationships (such as rules 2 and 4), since there is an indirect dependency
of Unique Pages with Pages). And there are also rules that show interesting relationships for edu-
cational purposes (such as rules 8, 12, and 13), which can be very useful for the instructor in deci-
sion making about the activities of their courses. It is proved that high grades, values greater than
6.9675, correspond to courses with high Quality (rule 6). Also, in rules 11, 12, and 13 there is a
composite dependency between the attributes Sessions and Grade. The last two remarks are use-
ful for an instructor, since he/she can pay more attention to the courses with low values of Quali-
ty.

Practical Implications
The application of data mining techniques proved that there is a relationship between course
usage and the corresponding student Grade. Additionally, rules 8, 12, and 13 offer to the instruc-
tors a lot of action ability, since they can pay more attention to the courses with low values of
Quality and Sessions. The instructor can be motivated to increase the quality and the quantity of
the educational material. More sessions means that more users (students) use the E-Learning plat-
form.
However there is a number of students who try to read the materials only just before the exams.
None deny that a good site with good material and which is updated frequently exists but rarely
visited. On the other hand, a bad website may have frequent visits because students’ visits are
related to their expected grade. So, the frequency and pattern of individual student accessing in-
structor’s materials will be an indicator to show such student’s “Laziness” or “Diligence”.
Feedback about the approach was received by the educators. The educators were informed about
the indexing results along with abstract directions on how to improve their courses. Most of them
increased the quality and the quantity of their educational material. They increased the quality by
reorganizing the educational material in a uniform, hierarchical, and structured way. They also
improved the quantity by embedding additional educational material. By updating educational
material, both quality and quantity were increased. A major outcome through the process of in-
forming the educators about the results is that the ranking of the courses constitutes an important
motivation for the educators to try to improve their educational material. Because of their mutual
competition, they want their courses to be highly ranked. A few educators complained that their
courses organization does not assist them to have high final scores in the ranking list. They ar-
gued that, for example, the metric interest is heavily influenced by the number of web pages used
to organize the educational material. Thus, courses that have all their educational material orga-
nized in a few pages have a low interest score. They were asked again to re-organize the material
for each course in the E-Learning according to the order they are taught, in order to facilitate use
by the students.
The fact that there were investigated only 39 courses in one platform only is a limitation for the
study. Especially, the data mining techniques demand larger datasets. However this was inelucta-
ble since case study department had this number of online courses.
In the future we intend to apply the same approach in other universities in other countries with
different cultures. In some different cultures the face-to face learning is preferable. It is remarka-
ble that e-Learning usually stands together with conventional learning and supplements it rather
than fully substituting it.

200
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

In some countries with strict hierarchical structure, people learn not to disagree with those who
are older than themselves. So, with E-Learning, a barrier may exist which modifies the interaction
between the teacher and the student.
From a pedagogical point of view this approach contributes to the improvement of course content
and course usability and the adaptation of the courses in accordance to student capabilities. Im-
provement of course quality gives students the opportunity of asynchronous study of courses with
actualized and optimal educational material with increased usage of the course material. Accord-
ing to our experiment, results usage is closely related to students’ grades. An increased usage
leads to better students’ grades and therefore to an improved educational outcome.

Conclusions
The proposed approach uses existing techniques in a different way to perform E-Learning usage
analysis. The metrics enrichment, homogeneity, and interest are used. Clustering, classification,
and association rule mining of the courses usage is presented. Two algorithms for course classifi-
cation and suggested actions are used.
It has the following advantages. (i) It is independent of a specific platform, since it is based on the
Apache log files and not the platform itself. Thus, it can be easily implemented for every E-
Learning platform. (ii) It uses indexes and metrics in order to facilitate the evaluation of each
course in the E-Learning and allows the instructors to make any necessary adjustments to their
course educational material. (iii) It applies data mining techniques to the E-Learning data. (iv)
One algorithm for the E-Learning data classification of the courses is used.
At present, the calculation of the metrics and the experiments of the algorithms are being generat-
ed manually. Therefore, some future work is needed to overcome such limitation. Thus, a plug-in
tool is being developed to automate the whole procedure. This tool will run in periodically (each
month) and will e-mail the instructors the results and the suggestions for their courses. A similar
policy was also applied by Feng and Heffernan (2005), where after long term observation the in-
structors were informed automatically by email about the quality of the content of their E-
Learning courses.
It should be mentioned that even if the scope of the method is on E-Learning platforms and edu-
cational content, it can be easily adopted by other web applications such as e-government, e-
commerce, e-banking, or blogs. Furthermore, Enrichment and Homogeneity metrics may also be
used, for example, by e-government applications, since enrichment shows how much information
is handed over to the end user and homogeneity characterizes the percentage of information inde-
pendently discovered by each user.

References
Agrawal R., Imielinski, T., & Swami, A. N. (1993). Mining association rules between sets of items in large
databases. Proceedings of SIGMOD (pp. 207-216).
Agrawal, R., Mannila, H., Srikant, R., Toivonen, H., & Verkamo A. (1996). Fast discovery of association
rules. In U. M. Fayyad, G. Piatetsky-Shapiro, P. Smyth, & R. Uthurusamy (Eds.), Advances in knowl-
edge discovery and data mining (pp. 307- 328). Menlo Park, CA: AAAI/MIT Press.
Agrawal, R., & Srikant, R. (1994). Fast algorithms for mining association rules. Proceedings of 20th Inter-
national Conference on Very Large Data Bases (pp. 487-499).
Avouris, N., Komis, V., Fiotakis, G., Margaritis, M., & Voyiatzaki, G. (2005). Logging of fingertip actions
is not enough for analysis of learning activities. Proceedings of Workshop Usage Analysis in learning
systems (AIED’05), Amsterdam.

201
E-Learning Platform Usage Analysis

Baker, R., Corbett, A., & Koedinger, K. (2004). Detecting student misuse of intelligent tutoring systems.
Proceedings of the Seventh International Conference on Intelligent Tutoring Systems, (pp.531–540).
Castro, F., Vellido, A., Nebot, A., & Minguillon, J. (2005). Detecting a typical student behaviour on an e-
learning system. In Simposio Nacional de Tecnologı´as de la Informacin y las Comunicaciones en la
Educacion (pp. 153–160). Spain: Granada.
Castro, F., Vellido, A., Nebot, A., & Mugica, F. (2007). Applying data mining techniques to e-learning
problems: A survey and state of the art. In L. C. Jain, R. Tedman, & D. Tedman (Eds.), Evolution of
teaching and learning paradigms in intelligent environment. Studies in Computational Intelligence 62,
Springer-Verlag.
Cone, J. W., & Robinson, D. G. (2001). The power of e-performance. Training & Development, 55(6), 32–
41.
Delacey, B., & Leonard, D. (2002). Case study on technology and distance in education at the Harvard
Business School. Educational Technology and Society, 5(2), 13-28.
Feng, M., & Heffernan, N.T. (2005). Informing teachers live about student learning: Reporting in the as-
sistment system. Proceedings of the 12th Annual Conference on Artificial Intelligence in Education
2005, Amsterdam.
Fichter, D. (2002). Intranets and e-learning: a perfect partnership. Online, 26(1), 68-71.
García, E., Romero, C., Ventura, S., & de Castro, C. (2008). An architecture for making recommendations
to courseware authors using association rule mining and collaborative filtering. Journal of User Model-
ing and User-Adapted Interaction, 19 (1-2), 99-132.
GUNet. (2011). Retrieved April 10, 2011 from http://eclass.gunet.gr/
Hamalainen, W., & Vinni, M. (2006). Comparison of machine learning methods for intelligent tutoring
systems. Procceedings of Int. Conf. in Intelligent Tutoring Systems, (pp. 525–534).
Holte, R. C. (1993). Very simple classification rules perform well on most commonly used datasets. Ma-
chine Learning, 11, 63-91.
Kaufmann, L., & Rousseeuw, P. J. (1990). Finding groups in data: An introduction to cluster analysis.
New York: John Wiley & Sons.
Kazanidis, I., Valsamidis, S., & Theodosiou, T. (2009). Proposed framework for data mining in e-learning:
The case of Open e-Class. Proceedings of Applied Computing 09, Rome, Italy.
Khribi, M. K., Jemni, M., & Nasraoui, O. (2009). Automatic recommendations for e-learning personaliza-
tion based on web usage mining techniques and information retrieval. Educational Technology & Soci-
ety, 12 (4), 30–42.
Kotsiantis, S. B., Pierrakeas, C. J. & Pintelas, P. E. (2003). Preventing student dropout in distance learning
using machine learning techniques. Proceedings of 7th International Conference on Knowledge-Based
Intelligent Information and Engineering Systems (KES), (pp 267–274).
Kotsiantis, S. B., Pierrakeas, C. J., & Pintelas, P. E. (2004). Predicting students’ performance in distance
learning using machine learning techniques, Applied Artificial Intelligence (AAI), 18(5), 411 – 426.
Koutri, M., Avouris, N., & Daskalaki, S. (2005). A survey on web usage mining techniques for web-based
adaptive hypermedia systems. In S. Y. Chen & G. D. Magoulas (Eds.), Adaptable and adaptive hyper-
media systems (pp. 125–149). IRM Press.
MacQueen, J. (1967). Some methods for classification and analysis of multivariate observations. In Pro-
ceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, ( pp. 281–297).
California, USA.
Mazza, R., & Dimitrova, V. (2007). CourseVis: A graphical student monitoring tool for supporting instruc-
tors in web-based distance courses. International Journal of Human-Computer Studies, 65(2), 125–
139.

202
 Valsamidis, Kontogiannis, Kazanidis, & Karakos

Mazza, R., & Milani, C. (2004). GISMO: A graphical interactive student monitoring tool for course man-
agement systems. Proceedings of International Conference on Technology Enhanced Learning ’04
(T.E.L.’04). Milan, Italy.
Minaei-Bidgoli, B., Kashy, D. A., Kortemeyer, G., & Punch, W. (2003). Predicting student performance:
An application of data mining methods with an educational web-based system. Proceedings of 33rd
Frontiers in Education Conference, (pp. T2A13–T2A18).
Minaei-Bidgoli, B., Tan, P-N, & Punch, W.F. (2004). Mining interesting contrast rules for a web-based
educational system. Proceedings of Int. Conf. on Machine Learning Applications, Louisville, USA
2004 (pp. 320- 327).
Mostow, J., Beck, J., Cen, H., Cuneo, A., Gouvea, E., & Heiner, C. (2005). An educational data mining
tool to browse tutor-student interactions: Time will tell!. Proceedings of workshop on educational data
mining (pp. 15–22).
Normark, O. R., Cetindamar, D. (2005) E-learning in a competitive firm setting. Innovations in Education
& Teaching International, 42(4), 325-335.
Pabarskaite, Z., & Raudys, A. (2007). A process of knowledge discovery from web log data: systematiza-
tion and critical review. Journal of Intelligent Information Systems, 28, 79-114.
Romero, C., Gutierez, S., Freire, M., & Ventura, S. (2008). Mining and visualizing visited trails in web-
based educational systems. In Educational Data Mining 2008, Proceedings of the 1st International
Conference on Educational Data Mining (pp. 182-186). Montreal, Quebec, Canada.
Romero, C. & Ventura, S. (2007). Educational Data Mining: a Survey from 1995 to 2005. Elsevier Journal
of Expert Systems with Applications, 33 (1), 135-146.
Radcliffe, D. (2002). Technological and pedagogical convergence between work-based and campus-based
learning. Educational Technology and Society, 5(2), 54-59.
Srinivasa, R. (2005). Data mining in e-commerce: A survey. Sadhana, 30(2 & 3), 275–289.
Starr, R. M. (1977) Delivering instruction on the World Wide Web: Overview and basic design principles.
Educational Technology, 37(3), 7-15.
Ueno, M. (2002). Learning-log database and data mining system for e-learning, Proceedings. of Interna-
tional Conference on Advanced Learning Technologies, ICALT 2002, (pp. 436-438).
Valsamidis, S., Kazanidis, I., Kontogiannis, S., & Karakos, A. (2010). Automated suggestions and course
ranking through web mining. Proceedings of 10th IEEE International Conference on Advanced Learn-
ing Technologies ICALT 2010, Sousse, Tunisia.
Valsamidis, S., Kontogiannis, S., Kazanidis, I., & Karakos, A. (2010). Homogeneity and enrichment: Two
metrics for web applications assessment. Proceedings of 14th Panhellenic Conference on Informatics.
Witten, I., & Frank, E., (2005). Data mining practical machine learning tools and techniques. San Francis-
co: Morgan Kaufmann.
Yan, T. W., Jacobsen, M., Garcia-Molina, H., & Dayal, U. (1996). From user access patterns to dynamic
hypertext linking. Proceedings of the Fifth International World Wide Web Conference on Computer
networks and ISDN systems, (pp. 1007-1014), Paris, France.
Zaiane, O. R. (2001). Web usage mining for a better web-based learning environment. Proceedings of Con-
ference on Advanced Technology for Education.
Zorrilla, M. E., & Álvarez, E. (2008). MATEP: Monitoring and analysis tool for e-learning platforms, Pro-
ceedings of the Eighth IEEE International Conference on Advanced Learning Technologies, (pp. 611-
613).

203
 E-Learning Platform Usage Analysis

Biographies
S. Valsamidis is a PhD candidate student at the Dept. of Electrical and
Computer Eng., Democritus University of Thrace, Xanthi, Greece. He
received a five-year Electrical Eng. diploma from Department of Elec-
trical Eng., University of Thessaloniki, Greece and MSc in Computer
Science from University of London, UK. He is an Applications Profes-
sor in the Dept. of Accountancy, Kavala Institute of Technology,
Greece. His research interests are in the areas of database and data
mining, data analysis and web applications assessment. His e-mail is:
svalsam at ee.duth.gr.

S. Kontogianns is a PhD candidate student at the Dept. of Electrical

and Computer Eng., Democritus University of Thrace, Xanthi, Greece.
He received a five-year Eng. diploma and MSc in Software Eng. from
Department of Electrical and Computer Eng., Democritus University
of Thrace. His research interests are in the areas of Distributed sys-
tems, middleware protocol design, network modelling and computer
networks performance evaluation. His e-mail is: skontog at ee.duth.gr.

Ioannis Kazanidis received the MSc degree in Computing from the

Coventry University and the PhD degree in Educational Technology
and Adaptive Educational Systems at University of Macedonia. Cur-
rently he is an adjoined assistant professor at Kavala Institute of Tech-
nology Greece. His main research interests lie in the area of adaptive
systems, data mining methods and algorithms, and computer supported
collaborative learning. He has published 6 papers in International Jour-
nals and Book chapters, and more than 12 papers in proceedings of
conferences.

A. Karakos received the Degree of Mathematician from the Depart-

ment of Mathematics from Aristotle University of Thessaloniki,
Greece and the Maitrise d' Informatique from the university PIERRE
ET MARIE CURIE, Paris. He completed his PhD studies at university
PIERRE ET MARIE. He is Assistant Professor at the Dept. of Electri-
cal and Computer Eng., Democritus University of Thrace, Greece. His
research interests are in the areas of Distributed systems, data analysis
and programming languages. His e-mail is: karakos at ee.duth.gr.

204

View publication stats