Education and Information Technologies

https://doi.org/10.1007/s10639-020-10185-5

Learner modeling in cloud computing

Sameh Ghallabi 1 & Fathi Essalmi 1 & Mohamed Jemni 1,2 & Kinshuk 3

Received: 10 February 2020 / Accepted: 7 April 2020/

# Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract
With the emergence of technology, the personalization of e-learning systems is en-
hanced. These systems use a set of parameters for personalizing courses. However, in
literature, these parameters are not based on classification and optimization algorithms
to implement them in the cloud. Cloud computing is a new model of computing where
standard and virtualized resources are provided as a service through the Internet. This
paper proposes an approach that allows learner modeling in the cloud where these
parameters are integrated. The suggested approach is based on the support vector
machine algorithm, which analyzes the learners’ traces to find the best classification
of learners through selected parameters with a low cost. An experimentation is con-
ducted to validate this approach. This experimentation is based on the produced traces
for learner modeling. The obtained results show that this approach represents the
learner model with low operation costs compared to classic systems (no cloud).

Keywords Learner modeling . personalized learning systems . learners' traces .

personalization parameters . SVM algorithm . Cloud computing

1 Introduction

The learner model is an essential component of personalized learning systems. It

contains personalization parameters that define the characteristics and needs of learners,
such as interests, preferences, and levels of knowledge, goals, tasks, backgrounds,

* Sameh Ghallabi
gallabisameh333@gmail.com

Fathi Essalmi
fathi.essalmi@isg.rnu.tn
Mohamed Jemni
mohamed.jemni@fst.rnu.tn
Kinshuk
kinshuk@ieee.org

Extended author information available on the last page of the article

 Education and Information Technologies

learning performances, learning styles, aptitudes and environments, as well as other

useful features (Aljohany et al. 2018). In order to achieve the personalization of e-
learning courses, learning systems use different personalization parameters. In the
literature, several approaches have used learner model in order to personalize learning
(Ahmed et al. 2017; Khamis 2015). The goal of personalization is to fit the learners’
needs and their learning goals (Jalal and Mahmood 2019; Alsadoon 2020).
The personalized learning systems limit the network and the space of their func-
tioning. There is no sharable or interoperable platform capable of integrating the
various existing systems. The users perform the mentioned systems from downloaded
software on their computers or from a physical server situated in their buildings.
Cloud computing presents a solution to meet such needs. It is a new technology that
allows users to access applications through the Internet. It is a combination of several
services: Software (SaaS), Platform (PaaS) and Infrastructure (IaaS) as a service. In
literature, a lot of work has considered the benefits of the cloud for applying this
technology in the e-learning personalization field. For instance, Zhao et al. (Zhao et al.
2017) suggested a new solution which used Google’s cloud computing platform and
Google App Engine to develop an open online learning platform. Their approach
utilized the pedagogical approach to provide learning concepts adapted to the learner’s
performance. Saini et al. (Saini et al. 2017) presented a new solution that integrated an
e-learning system into the cloud. Their approach highlighted the learning resources
which had to be appropriate for each learner according to the media preference.
Another example is Aggarwal et al. (Aggarwal et al. 2017) who put forward a new
approach that allowed building a personalized learning environment for education.
However, very little research has been available which focuses on learner modeling in
the cloud. Added to that, the personalization parameters have not been based on
classification and optimization algorithms to implement them in the cloud. This leads
to the following question: How can we realize learner modeling in cloud computing
where these parameters are integrated with low operation costs? To answer this
question, this paper presents an approach that uses the cloud computing for learner
modeling through the selected personalization parameter.
Each parameter is represented as a Web service. Hence, this research offers stan-
dardized, reusable and shared parameters which are delivered over the Internet, and
therefore accessible and available anywhere. Furthermore, this approach allows
representing the learner model in a flexible, reliable, accessible and available way. It
which provides better storage space and virtualization of personalization parameters.
Learner modeling is structured utilizing cloud. Thus, it is available to any user
application that requests it. The gathered information can be utilized in other learning
tools. Through this approach, teachers access learner models in order to create a course
adapted to their needs. In addition, they look for personalization parameters to access or
use them as a basis of the course that they wish to build.
The proposed solution is based on the Support Vector Machine (SVM) algorithm
(Wiyono and Abidin 2019), which uses and analyzes the learners’ traces to extract the
most optimal and best classification of learners through the selected personalization
parameter. It utilizes this algorithm to give optimal results, more accuracy and a high
performance. In (Mohamed 2017), the SVM was compared with the most supervised
machine learning algorithms for classification, such as the decision tree, artificial-
neural-network and K-nearest-neighbor (see section 3).
Education and Information Technologies

The rest of the paper is organized as follows. Section 2 defines cloud computing and
describes the benefits of cloud computing for the personalization of e-learning. It
outlines some related work about learner modeling. Section 3 presents the proposed
approach which allows learner modeling in a cloud environment. In section 4, the
suggested approach is explained with a small example. Section 5 presents the discus-
sion. Finally, conclusion and future work are presented in section 6.

2 Related works

This section is divided into two subsequent sections. The first one defines the cloud
computing environment and the three service models (SaaS, PaaS and IaaS). The
second one presents related work regarding personalization in the cloud.

2.1 Definition of cloud computing

Cloud computing is a new paradigm that uses Internet to maintain data and applications
(Rad et al. 2018). It provides computing resources that are delivered as services over a
network. Then, these resources are represented in a standardized and shared way. Users
have the possibility to invoke these services by computing devices anywhere and
anytime. Cloud computing offers three service models that can be grouped into the
following categories:
SaaS: It offers software applications as services on the internet. It offers ready
operational software for use, without passing through a stage of installation and a task
of maintenance. The applications are accessible to users through a Web browser. It is an
invocation of services from a service provider.
PaaS: It is found at the level above SaaS. It can be used to execute applications. PaaS
provider gives subscribers access to the resources they need to develop. They also
operate the applications over the Internet. Thus, the client maintains its own applica-
tions while the provider maintains servers and software infrastructure (databases,
security, and storage). Hence, in order to develop, test and run their application, a user
must to rent a platform.
IaaS: It is the infrastructure part of the cloud. Only the server is dematerialized.
Providers of IaaS offer computers-physical virtual machines and other resources. IaaS
allows users to dispose infrastructure on demand. Users of IaaS can therefore use
virtual servers on demand situated in data centers. Through such infrastructure, the
users can accommodate and execute applications or store data.
To sum up, personalized learning systems use several components in order to create
a course adapted to learners’ needs and characteristics.
In this section, the advantage of using cloud computing for learner modeling
according to selected e-learning personalization parameters are presented. There are
numerous advantages when the personalized learning systems are implemented with
the cloud computing. The cloud allows for greater flexibility and mobility in the use of
resources for teaching and learning (Divya and Prakasam 2015). There is no hardware
or software to install and to buy. The learning contents should be made available and
accessible to the teachers and students. The data or software applications are not stored
on the user’s computer, but rather are accessed through the web from any device at
 Education and Information Technologies

anywhere. In addition, the cloud allows users to store almost all kinds of content and
data including applications, documents, music, photos and much more.
To compare between personalized learning systems in cloud and classical systems (a
classical system is the personalization of e-learning which provides learning contents
adapted to the learners’ needs and characteristics), seven criteria of comparison, found
in [], are presented as follows:

Reduced costs This cost is a set of expenses supported by a company. Users execute
the classical personalized learning systems from software downloaded on a computer or
a physical server situated in their buildings. However, they can run systems from the
cloud through their mobile phones or PCs having minimum configuration with Internet
connectivity.Therefore, in order to create the educational resources, users do not need
to spend more money for large memory data storage in local machines.

Anytime_Anywhere This criterion allows teachers to access their required resources

over the internet anywhere and anytime. In cloud, the Internet connection must be of
sufficiently high availability and performance for that particular service.

Availability Availability is used to guarantee a permanent service. Cloud computing

provides permanent services with the use of redundant systems.

Networks This criterion allows for providing computing resources over the internet.
Cloud based personalized learning systems eliminate time restrictions of the network
and space. On the other hand, classic personalized learning systems have limited
network and space for their operations.

Storage capacity This criterion relates to the storage space of learning resources. Cloud
computing supports multi-tenant usage, where cloud service provider makes the
necessary resources available to multiple users using technologies.

Scalability This criterion relates to the system’s capacity to process a large number of
operations or transactions during a given period without deteriorating performance.
Cloud resources can be dynamically delivered in real-time. Cloud computing provides
pool of large amount of resources from data centers and makes them easily accessible.
On the other hand, classic personalized learning systems are characterized by their
static and manual scalability.

Run time This criterion relates to the time taken by the systems to run the user’s
request. Cloud based personalized learning systems can respond instantly to the user’s
demand, due to the high capacity of remote servers. On the other hand, the run time for
classic personalized learning system service becomes larger when growth of the
numbers of learning actors grows.
Table 1 summarizes the above mentioned systems and shows the various criteria
used to compare between cloud based personalized learning systems and classical
systems. The columns represent the two types of learning systems, and the rows list
various criteria. Each cell includes a notation representing the importance of the
criterion (represented in that row) in the learning system (represented in that column).
Education and Information Technologies

Notation *** means that the criterion is fully considered, ** means considered little, *
means considered rarely, and – means the criterion is not considered.
There are a lot of advantages from using cloud computing for personalized learning
systems. In the cloud, users store their data, and information usually requires authen-
tication. It enables students to access the learning objects that suit their characteristics
and needs using any device. In addition, it provides software, hardware and a good
platform to enhance e-learning personalization services. The authors in (Sasikumar and
Saravanakumar 2017; Pourqasem et al. 2014) compared and presented the difference
between cloud computing and other technologies. Table 2 shows the comparison
between cloud computing and a simple PC through the mentioned criteria.
To sum up, cloud computing can solve some of the challenges of personalized
learning systems. It is characterized by its scalability, efficiency, flexibility, and
reliability. It offers an unlimited learning choice for students. Therefore, thanks to its
promising benefits, many works have been done in order to apply them in the e-
learning personalization field.

2.2 Application of personalization in the cloud computing

Various technologies have emerged in recent years that have enhanced e-learning person-
alization. One new direction is cloud computing. Cloud computing is an emerging new
computing model. It creates virtualized resources which can be made available to users.
Several works have been done in order to integrate of cloud computing to meet the
requirements for personalized learning systems. For example, Jeong and Hong (Sasikumar
and Saravanakumar 2017) proposed a service based personalized learning system in cloud
computing technology. Their approach provided learning process according to the learners’
level of knowledge. Moreover, Kim’s work (Jeong and Hong 2012) propounded a cloud-
computing-based Adaptive Education Hypermedia System (AEHS). The AEHS allows
creating the learning content according to the motivation level and the media preference. In
addition, Palanivel and Kuppuswami (Kim et al. 2013) suggested new architecture to
personalize e-learning systems using service-oriented cloud computing. Furthermore,
Yadav’ work (Palanivel and Kuppuswami 2014) presented an approach to integrate per-
sonalization e-learning systems into cloud environment. Their approach allowed students to

Table 1 Comparison between two types of systems []

Learning systems Classic personalized Cloud personalized

Different criteria learning systems learning systems

Low cost ** ***

Run time * ***
Networks * ***
Scalability – ***
Mobility and decentralization – ***
Availability ** ***
Storage capacity * ***
 Education and Information Technologies

Table 2 Comparison between various technologies

Technologies Different criteria Cloud computing Simple Pc

Cost Pay per use Undefined business model

Run time Fast Slow
Networks Internet Undefined
Scalability High because it is real-time Low because of static and
manual services.
Mobile and decentralized Decentralized resources Centralized resources
Availability High Low
Storage capacity High capacity Low capacity

access a wide array of resources and software tools that suit their learning styles and interests.
Another example is Bosamia and Patel (Yadav 2014), who proposed a new approach which
allowed the application of cloud computing in e-learning. The goal of their approach was to
provide a suitable learning objects according to the media preference. Further, Laeep and
Shaikh (Bosamia and Patel 2016) investigated the numerous opportunities for adopting
cloud computing in an e-learning environment. Their approach utilized cloud to secure data
storage and computing power. Jain and Chawla (Laeep and Shaikh 2016) presented a new
approach that allows the application of cloud computing in e-learning. Further, Anand and
Kamayani (Jain and Chawla 2013) defined the advantages of cloud computing for e-
learning systems. The main objective of the work was to provide a shared pool of the e-
learning resources. Finally, Rad et al. (Anand 2015) proposed to open an interactive cloud
based learning platform called Cloud-eLab. This platform provided personalized learning
based on automatic cognitive feedback for each learner with flexibility to repeat the
experiments at their own pace.
Personalized learning systems use and combine a set of personalization parameters for
personalizing courses. Previous work in the literature has presented different parameters. For
example, ASSA (Aljohany et al. 2018) allowed to generate the appropriate questions in an
adaptive way according to the learner’s level of knowledge, the learning goals and the
learning style. Another example is AES (Siddique et al. 2018), who used the Felder
silverman learning style to offer learning courses adapted to the learners’ needs. Besides,
Harandi’ work (Khamis 2015) presented an approach which permitted providing a person-
alized learning object according to the motivation level.
Table 3 represents the availability of personalization parameters. Availability is a
criterion which allows providing permanent personalization parameters. This table
shows the availability of personalization parameters which are defined in columns. It
presents the reported learning systems and personalization parameters in lines. Each
cell includes a notation representing the availability of these parameters. Notation*
means that the parameter is available in the Web, ** means it is available in the cloud,
and – means it is not available.
As shown Table 3, the personalized learning systems use several parameters in order to
generate a learning objects adapted to the profiles of learners and to the specifics of courses.
We observe that most of the personalization parameters are not accessible and not available
Education and Information Technologies

Table 3 Classification of personalization parameters through their availability in Web and/or in Cloud

Reported personalization Personalization Availability of personalization

systems parameters parameters

Available Available in cloud

in Web

(Aljohany et al. 2018; Learners’ level of * –

Anand 2015), FORMATION PATH (Siddique Knowledge
et al. 2018), AEHS-LS (Kumar et al. 2013) (LK)
(Anand 2015), AEHS-LS (Kumar et al. 2013) Media Preference * **in the
(MP) SOCCAPES
(Jeong and Hong
2012) 2014) and
in
the cloud AEHS
(Pourqasem et al.
2014)
(Riad et al. 2012; Eltigani Ali Mustafa and Kolb Learning * –
Mohamed Sharif 2011) Cycle (KLC)
Motivation Level * ** in the cloud
(ML) AEHS
(Pourqasem et al.
2014)
(Gholipour et al. 2011; Cognitive traits * –
Gholipour et al. 2011), AMDPC (CT)
(Virvou and Troussas 2011;
Behaz and Djoudi 2012;
Saleena and Srivatsa 2015)
(Alsadoon 2020) Language * –
Preference (LP)
(Gholipour et al. 2011) Honey and * -
Mumford * -
learning
Style (HMLS)
(Lazarinis et al. 2010) Learning Goals * –
(LG)
(Virvou and Troussas 2011) La Garanderie * –
Learning Style
(GLS)
(Alsadoon 2020) Pedagogical * –
approach
(Jain and Chawla 2013; Anand 2015; Felder–Silverman * -
Dwi Surjono, 2014) Learning style * -
FORMATION PATH (Siddique et al. 2018;
Gholipour et al. 2011), Moodle –LS System
(Ali Salahli et al. 2012), AEHS-LS (Kumar et al.
2013), ADMCA (Limongelli et al. 2011), AES
(Karadimce and Davcev 2012)

in the cloud environment. In addition, these parameters are not based on classification and
optimization algorithms to implement and integrate them in the cloud.
 Education and Information Technologies

The research presented in this paper defines a new solution which takes into account
the benefits of cloud computing for learner modeling according to the selected person-
alization parameters at low operation costs and time. The proposed solution is based on
the SVM algorithm, which is presented in the next section.

3 Proposed approach

3.1 Research methodology

This section presents the advantages of using both cloud computing and the SVM
algorithm for learner modeling. As shown Table 2, the various benefits of cloud
computing provide a flexible and enriched environment. In addition, this paper attempts
to apply this technology so as to maximize the availability and high storage space of
personalization parameters. These parameters are given as a service through the
Internet. This technology allows representing the learner model in a reliable, reusable,
accessible and available way. It also permits the interoperability between heterogeneous
information resources and services. It enables teachers to use this model without
installing any system, and to access these parameters anywhere and anytime.
On the other hand, this paper uses the SVM algorithm which analyzes traces about
learners to find the best classification of learners through the selected personalization
parameters. There are several measures for evaluating the classifier performance. In
(Wainer 2016), the SVM is compared with the most supervised machine learning algorithms
for classification decision tree, artificial-neural-network, and K-nearest-neighbor. The ob-
tained results in (Wainer 2016) has shown that the SVM had the best prediction accuracy
with a precision value of 95%. Whereas, the decision tree algorithm had 93% prediction
accuracy and the K-nearest-neighbor algorithm had a prediction accuracy value of 92%. In
addition, Wainer’ work (Atanasov and Ruskov 2016) demonstrated that SVM could
successfully deal with complex classification problems. Besides, the authors in (Aggarwal
et al. 2017) used four criteria to compare between the SVM and the mentioned algorithms.
These criteria are presented as follows:

& Accuracy: It is the measure of how correctly the algorithm classifies the instances.
& Sensitivity: It measures the proportion of correctly classified positive examples.
& Specificity: It measures the proportion of correctly classified negative examples.
& Kappa: It measures the performance of an algorithm.

Table 4 presents the comparison between the classification of algorithms according to

the previous criteria.
To sum up, the SVM has the best accuracy and the highest kappa amongst all
classifiers. There are a lot of advantages of the SVM algorithm. This algorithm
provides very fast and memory- efficient implementations. It has a high generalization
property for the generated classifier model. The goal of this algorithm is the search for
the best hyperplane separating the data classes. This algorithm allows giving more
accuracy, a higher performance and reliability. It is the most important method which is
applied in various areas such as data mining, bioinformatics, and image processing. In
addition, it is applicable for real time applications. Furthermore, it is used as a tool to
Education and Information Technologies

Table 4 Comparison between learning machine algorithms (Mohamed 2017)

Criteria Accuracy Specificity Sensitivity Kappa

Algorithms

SVM 0.763 0.487 0.881 0.3948

Artificial neural network 0.709 0.507 0.796 0.3038
Decision tree 0.739 0.487 0.847 0.3495
K-nearest-neighbor 0.713 0.253 0.81 0.1929

solve supervised learning classification problems. It has a regularization parameter. It is

defined by a convex optimization problem. The SVM provides the largest flexibility in
the choice of the form of the threshold. It permits finding the best classifier and having
better performances practically in most cases. Besides, it enables analyzing a large
amount of data to identify the optimal classification. For this reason, this paper adopts
the mentioned algorithm as the best solution to analyze learners’ traces in order to
extract the best and most reliable classification through the selected personalization
parameter. It utilized this algorithm to give optimal results and a good performance.

3.2 Integration of personalization parameters in cloud

This paper proposes a new approach for learner modeling in the cloud computing
environment. This learner model is delivered over the Internet, so it is accessible and
available anywhere and anytime. The suggested approach aims to enhance the repre-
sentation of information about learners and to integrate the personalization parameters
in the cloud with low operation costs. These parameters are considered as a collection
of Web services which are going to be used according to cloud computing. All services
are uploaded to the databases.
Teachers can access these databases to select and choose the best and most
optimal parameter they want to use. If a teacher does not find the parameter they
are looking for, they can conduct a search by using an engine of discovery. They
have the opportunity to quickly and economically access the learner model
through any device (such as tablets and phones) anywhere and anytime. The
proposed approach allows representing these parameters in a flexible, standard-
ized, shared, accessible and available way. The architecture of this approach is
composed of four main components (learner interface, learner’ traces, SVM
algorithm and learner modeling), as illustrated in Fig. 1.

SVM algorithm It is a set supervised learning techniques that are used for both
classification and regression challenges (Wainer 2016). It is mostly used to solve
classification problems. It analyzes the learners’ traces to draw out the best and most
reliable classification through the selected personalization parameter.

Learner interface It allows to collecting the learner ‘s behaviors.

Dataset learner’ traces They present the traces about students while learning.
 Education and Information Technologies

Fig. 1 Learner modeling in cloud environment

Learner modeling It contains personalization parameters. These parameters are a set of

learner characteristics and needs which can be used for the personalization of courses.
In particular, this paper presents two parameters: levels of knowledge and language
preferences. Learner’s level of knowledge is the subset of expertise’ s knowledge. It is
related to the learner ‘s background and previous knowledge when communicating
learning materials to the students. Every learner is characterized by three types of
learner ‘s values: beginner, intermediate and advanced. Language preferences provide
the student with the language of learning materials they prefer. For example, the values
of this parameter are French and English. Each parameter can be considered as a service
which is executed when it is called.
As depicted graphically in Fig. 1 (see Fig.1), this algorithm analyzes and uses
learners’ traces to classify them. This approach aims to integrate two parameters (level
of knowledge and language preference) in the cloud with low operation costs. These
parameters are delivered over the Internet and are therefore accessible anywhere.

4 Simulation

This section defines the different steps to validate the suggested approach. It presents
the procedure followed during the learners’ classification through the selected person-
alization parameter. Besides, it presents the obtained results which show that this
Education and Information Technologies

approach implements and integrates the learner model with low operation costs com-
pared to the classic systems (no cloud).

4.1 Procedure

The types of the students’ traces are essential for learner modeling. The proposed approach
focuses on test traces. Learners are classified according to their level of knowledge. This
approach uses traces (Spent time: type time, obtained scores: type number, learning
activities: type char, test: type char, and answers to questions: type integer) obtained from
36 learners who participated in the experiment which was done at a university in Tunisia.
Each student treats a test composed of 10 questions with multiple choices.
Furthermore, this approach uses the produced traces (Spent time: type time, choices
made: type char, learning activities: type char and answers to questions: type integer)
for learner modeling according to the language preference. We increase the number of
students to show the importance of the proposed approach.
This section presents a series of steps for extracting the best and most optimal
classification of learners through the selected personalization parameter. During the
learning process, many test traces are collected. In this approach, the dataset presents
the test traces generated by learners while learning. Two instruments are used in
simulation. In the first instrument, the SVM algorithm is utilized to analyze the five
traces, as presented above, to extract the best classification of learners in three classes
(beginner, intermediate and advanced) through the level of knowledge. In the second
instrument, the SVM algorithm uses four traces, as defined previously, to classify
learners according to the language preference. Learners are classified into two classes
(French and English) based on their choices.

4.2 Obtained results

To validate the proposed solution, we use the following tools:

CloudSim (Wexbridge and Nyland 2014):It is a simulator that allows modeling and
simulating cloud computing infrastructure and services. It enables developers to test the
performance of their application in a repeatable and controllable environment, free of
cost. The principal advantages of CloudSim are flexibility of defining configurations,
time efficiency and ease of use and customization.
Netbeans (Paavolainen 2016): It is an open source and is written in Java program-
ming. It is an Integrated Development Environment (IDE) that enables you to develop
desktop, mobile and Web applications. It includes user interface functions, source code
editors and version control. The IDE supports application development in various
languages (e.g. Java and PHP).
This section presents the different results of learner modeling in the cloud and in classical
systems. The SVM algorithm uses students’ traces to obtain the best and most optimal
classification of learners according to the selected personalization parameter. Learners are
classified into three classes in both pre and post_ tests based on their scores. These classes are:

Class 1_ Beginner: If the scores are less than 10 out of 20.

Class 2_Intermediate: If the scores are greater than or equal to 10 and strictly less
than 13.
 Education and Information Technologies

Class 3_Advanced: If the scores are greater than or equal to 13.

For example, Fig. 2 represents students’ classification according to their level of

knowledge in three classes (beginner, intermediate, advanced).
Learners are classified into two classes when answering questions based on their
choices. These classes are: class1_ language preference _ English (76.26%) and
class2_language preference _French (23.74%).
The goal of this approach is learner modeling with in short time and with a low cost.
To show the importance of this paper, we define a set of criteria to compare the
implementation of these parameters in two types of systems (cloud and classical
systems):

4.2.1 Availability

In the experimentation part, learners ‘traces are copied to two data centers. We use two
virtual machines of different characteristics. If the first data center is on load, this
approach passes automatically to the second data center. It also performs personaliza-
tion parameters as a Web service. This service is hosted on two different servers (e.g.
the widfly Runtime 9.0.1). The proposed approach allows users to access the person-
alization parameter when they want to using service searching. If the parameter is not
accessible in the server, then the other server responds automatically. In the literature,
the availability of cloud services has been contractually committed to ensure 99.9%
(e.g. 99.99% during work days, 99.9% for nights/weekends) (Elbeh and Biundo 2012).

Class 1_ Beginner: beg

Class 2_Intermediate: inter
Class 3_Advanced: adv
Fig. 2 Students’ classification according to the level of knowledge
Education and Information Technologies

4.2.2 Anytime_Anywhere

In this experimentation, each parameter (level of knowledge/ language preference)

is represented as Web service REST. (A Web service REST is a program hosted
on a server containing functions accessible via HTTP requests). This service is
developed with j2EE WS and accommodated on Widfly Runtine 9.0.1. All
services of these parameters are uploaded to the data centers. These parameters
are delivered over the Internet. The call of a parameter is realized via a resource
(URL). Thus, the developed parameters are accessible and available anywhere and
anytime. Teachers can access the data centers to choose and select their desired
parameter according to the profiles of learners and the specifics of courses. If a
teacher does not find the parameter they are looking for, they can conduct a search
by using an engine of discovery.

4.2.3 Time and cost

This approach takes into consideration both “Time” and “cost” criteria to compare
between learner modeling in cloud and classical environments. Tables 5 and 6
present the required time for learner modeling according to the learner’s level of
knowledge and the language preference. In this experiment, we note here that we
use the same characteristics of test data (Numbers of learners).
According to Tables 5 and 6, we observe that if the number of learners
grows, then their traces increase, and so does the execution time. In this
context, the proposed approach uses the execution-time criterion which allows
identifying the best environment. A cloud- based personalized learning system
permits constructing the learner model in short time. Whereas, with the growth
of learner’ traces, the execution time for a classic personalized learning service
is slow.

4.2.4 Cost

This approach offers the opportunity to represent the learner model with a low cost. As
shown in Tables 5 and 6, the time of learner modeling is decreased. Consequently, the
cost automatically declines.

Table 5 Execution time of level of knowledge in classical and Cloud environments

Environments Numbers of learners Execution time (s)

Classical Cloud

100 19 8
200 68 17
300 155 24
400 320 33
 Education and Information Technologies

Table 6 Execution time of language preference in classical and Cloud environments

Environments Numbers of learners Execution time (s)

Classical Cloud

100 17 6
200 62 13
300 140 19
400 310 28

5 Discussion

In the literature, several approaches have defined a lot of mechanisms used to create
and generate learner model. Alternatives A.1 and A.2 are discussed. Furthermore, A.1
is divided into two alternatives, named A.1.1 and A.1.2. Moreover, A.2 is divided into
two alternatives, which are A.2.1 and A.2.2.

5.1 Learner modeling in the web

Previous work has utilized different techniques (Aljohany et al. 2018; Ahmed et al.
2017) to create and generate an interoperable and reliable learner model. In addition,
classic personalized systems (Yang et al. 2013; Paavolainen 2016) may represent the
learner model in a reusable way. However, there have been several approaches in the
literature (International Journal of Mathematics, Science and Technology Education,
n.d.; Gutiérrez et al. 2016) which have not represented the learner model in an
interoperable and reusable way. As a consequence, the used learner model in a given
system is not exploited in other systems. Furthermore, learner models in the Web allow
storing the collection information about learners in a relevant and secure way. Classic
personalized learning systems have limited the storage space. In this case, users need to
spend much money for large memory data storage in local machines. As a result, the
creation of learner models is a very costly-operation. In addition, these models are
centralized and are not mobile.

5.1.1 Learner modeling based on SVM algorithm

The SVM algorithm allows constructing the student model from multiple behaviors.
This algorithm uses and analyzes traces about learners to find the best and most reliable
classification of learners. Moreover, it gives optimal results and a good performance.
With a growing number of learner’ traces, the execution time of the SVM algorithm is
slow as it is illustrated in Tables 5 and 6.

5.1.2 Other algorithms used for learner modeling

In the literature, personalized learning systems (Taniguchi et al. 2015; Rahadian and
Budiningsih 2017) have used different personalization algorithms (such as the
interbook, genetic and ant-colony algorithms) to adapt the learning contents according
Education and Information Technologies

to learners’ needs and characteristics. These algorithms are adaptation algorithms and
are not used to solve the classification problems.

5.2 Learner modeling in the cloud

Learner modeling in the cloud helps the ones interested in this field reduce the
programming effort and the cost of large memory data storage. A cloud based
personalized learning system provides better storage space and virtualization of per-
sonalization parameters. Each learner model is available to any user application that
requests it. Learner modeling is structured using cloud computing. The information
gathered can be utilized in other learning tools. Hence, it is represented in a reusable
and interoperable manner. In fact, the user can access this model regardless of their
location or computing device/ platform they use. Cloud providers offer various param-
eters that are accessible and available over the Internet anywhere and anytime. How-
ever, the problem of networks is treated by other researchers.

5.2.1 Learner modeling based on SVM algorithm

The produced traces about learners during the learning step are very rich files
(Ollagnier-Beldame and Mille 2012). In that case, using the selected personalization
parameters, the SVM algorithm analyzes a huge number of students’ traces to find the
best and most optimal classification of learners with the lowest cost and least time (see
Tables 5 and 6). This algorithm in the cloud allows identifying more accurate and
efficient students’ classifications.

5.2.2 Other algorithms used for learner modeling

The existing personalization algorithms in a cloud are genetic and SVM algorithms.
Personalized learning systems have used the genetic algorithm to adapt learning objects
according to learners’ profiles and needs. Nevertheless, this algorithm is not used to
solve classification problems.

6 Conclusions and future directions

This paper has proposed a new solution to improve learner modeling within personalized
learning systems. The goal of this solution is to construct a student model according to the
selected parameter in a flexible, reliable and reusable way. Besides, it allows implementing
and integrating the personalization parameters in the cloud with low operation costs and
time. Each parameter is represented as a Web service. Hence, these parameters are delivered
over the internet and therefore accessible and available anywhere and anytime. In this paper,
all services of these parameters are uploaded to the databases. Teachers can access the
databases to choose and select the best parameter that matches the characteristics of learners
and the specifics of courses. The proposed approach is based on the SVM algorithm which
analyzes and classifies traces about learners to extract the best and most reliable of learners’
classification. This algorithm permits giving optimal results. To show the feasibility and
effectiveness of this approach, a set of experiments has been conducted. The obtained results
 Education and Information Technologies

have enabled learner modeling with low operation costs and a good performance compared
to classic systems (no cloud). In addition, this work defines a set of criteria (such as:
operation cost, time, availability, and anytime_anywhere) to compare the generation of
learner model in two types of systems (Cloud-based personalized learning systems and
classical systems).
In future work, we will ameliorate the proposed approach by treating the following
preoccupation: (i: Choice of best parameter): In this paper, all parameters are accessible and
available as services and are uploaded to the databases. Teachers can access the databases to
choose the required parameter they want to use. However, this paper is needed to present
how to choose the best parameter according to the learners’ characteristics and the specifics
of the courses. Therefore, to overcome this limitation, we envision to integrate an adaptive
system which can help teachers select the best parameter from all candidate parameters with
low operation costs, or to use another mechanism (e.g. middleware, or service Web) to
choose the parameter that will satisfy the tutor’s request. In this case, the adaptive system and
another mechanism will treat the teacher’s request to give the best parameter through
learners’ needs and characteristics. (ii: Suspended connection problem): From a user’s point
of view, we can’t deny that cloud computing has drawbacks like the problem of suspended
connection (non-availability of services) which should be considered before taking advan-
tage of the cloud. To solve this problem, for example, we think that using the backup
technique can avoid losing data. (iii: Composition of personalization parameters): This paper
aims to integrate the afore-mentioned parameters in the cloud. Yet, this research is needed to
identify how to integrate the cloud computing technology so as to combine the best
personalization parameters with low operation costs. Then to solve this limitation, we project
using the composition process (for example, it includes composition operators, algorithms,
and so on) which will involve building a new parameter by combining the available and
existing parameters. The composition process will utilize a set of criteria (e.g: response time,
availability, etc) to select the optimum composition which will satisfy the tutor’s request.
Hence, it offers the opportunity to combine personalization parameters with low costs and
time. (iv: Security of learners ‘traces): Finally, this research is needed to identify how to
integrate the security aspect to keep learners’ traces. To solve this problem, for example, we
think that using the security technique can protect privacy traces.

Acknowledgements Not applicable.

Authors’ contributions Each author contributed evenly to this paper. All authors read and approved the
final manuscript.

Funding information Not applicable.

Compliance with ethical standards

Competing interests The authors declare that they have no competing interests.

Availability of data and material The datasets generated and/or analyzed during the current study are not
publicly available due to privacy reasons but are available from the corresponding author on reasonable
request.
Education and Information Technologies

References

Aggarwal, M., Rinkey and Gupta, P.” Impact of Cloud Computing in E-Learning: A Study,” International
Journal on Recent and Innovation 97 Trends in Computing and Communication, 5, 912–922 (2017).
Ahmed, M. U., Sangi, N. A., & Mahmood, A. (2017). A learner model for adaptable e-learning. International
Journal of Advanced Computer Science and Applications, 8, 139–147.
Ali Salahli, M., Ozdemir, M., & Yasar, C. (2012). Building a fuzzy knowledge management system for
personalized elearning. Procedia - Social and Behavioral Sciences, 46(2012), 1978–1982.
Aljohany, D. A., Salama, R. M., & Saleh, M. (2018). ASSA: Adaptive E-learning smart students assessment
model. International Journal of Advanced Computer Science and Applications, 9, 128–136.
Alsadoon, E. (2020). The impact of an adaptive e-course on students’ achievements based on the students’
prior knowledge. International Journal of Education and Information Technologies, 25, 1–11.
Anand, H. S. & Kamayani (2015). Scope of cloud computing in education sector: A review. International
Journal of Recent Research Aspects, 2(2), 150–152.
Atanasov, D., & Ruskov, T. (2016). Simulation of cloud computing environments with CloudSim. Journal of
Information Technologies and Control, 12(3–4), 1–6.
Behaz, A., & Djoudi, M. (2012). Adaptation of learning resources based on the MBTI theory of psychological
types. International Journal of Computer Science Issues, 9, 135–141.
Bosamia, M., & Patel, A. (2016). An overview of cloud computing for E-learning with its key benefits.
International Journal of Information Sciences and Techniques (IJIST), 6(1/2), 1–10.
Divya, P., & Prakasam, S. (2015). Effectiveness of cloud based E-learning system (ECBELS). Journal of
Computer Applications, 119(6), 29–36.
Dwi Surjono, H. (2014). The evaluation of a Moodle based adaptive e-learning system. Journal of Information
and Education Technology, 4(1), 89–92.
Elbeh, H., & Biundo, S. (2012). A personalized course generation system based on task centered instruction
strategy. International conference on artificial intelligence (ICAI).
Eltigani Ali Mustafa, Y., & Mohamed Sharif, S. (2011). An approach to adaptive ELearning hypermedia
system based on learning styles (AEHS-LS): Implementation and evaluation. Journal of Library and
Information Science, 3(1), 15–28.
Gholipour, A., Mazaheri, A., Haghshenas, E., & Tavakoli, M. (2011). An intelligent method for customizable
adaptive learning content generation. Journal of Information and Communication Technology Research,
3(1), 43–55.
Gutiérrez, I., Álvarez, V., Paule, M. P., Pérez-Pérez, J. R., & de Freitas, S. (2016). Adaptation in E-learning
content specifications with dynamic sharable objects. International Journal of Adaptive Educational
Technology Systems, 4(24), 1–11.
Jain, A., & Chawla, S. (2013). E-learning in the cloud. Journal of latest research in. Science and Technology,
2(1), 478–481.
Jalal, A., & Mahmood, M. (2019). Students’ behavior mining in e-learning environment using cognitive
processes with information technologies. International Journal of Education and Information
Technologies, 24, 2797–2821.
Jeong, H.Y., & Hong, B.(2012). Service based personalized learning system in cloud computing environment.
AST/EEC/MMHS/AIA conference program, 95-102.
Karadimce, A., & Davcev, D. (2012). Personalized multimedia content generation using the QoE metrics in
distance learning systems. In International conference on adaptive and self-adaptive systems and
applications (pp. 1–6).
Khamis, M. A. (2015). Adaptive e-learning environment systems and technologies. The First International
Conference of the Faculty of Education: Albaha University.
Kim, J., Byun, J., & Jeong, H. (2013). Cloud AEHS: Advanced learning system using user preferences.
Journal of Convergence, 4(3), 31–36.
Kumar, A., Nalavade, J. E., Yeola, V., Vivek, V., & Srivastava, Y. (2013). An adaptive learning system based
on ant Colony algorithm. Journal of Soft Computing and Engineering (IJSCE), 3(2), 212–214.
Laeep, K., & Shaikh, Z. A. (2016). Challenges and opportunities of cloud -based E-learning systems.
International Journal of Educational and Pedagogical Sciences, 10(2), 693–697.
Lazarinis, F., Green, S., & Pearson, E. (2010). Creating personalized assessments based on learner knowledge
and objectives in a hypermedia web testing application. Computers and Education, 55, 1732–1743.
Limongelli, C., Sciarrone, F., & Vaste, G. (2011). Personalized e-learning in Moodle: The Moodle_LS
system. Journal of e-Learning and Knowledge Society, 7(1), 49–58.
 Education and Information Technologies

Mohamed, A. E. (2017). Comparative study of four supervised machine learning techniques for classification.
International Journal of Applied Science and Technology, 7(2), 5.17.
Ollagnier-Beldame, M., & Mille, A. (2012). E-learning : Tracer l’activité des apprenants pour favoriser leur
appropriation des systèmes ? Revue du Management Technologique.
Paavolainen, S. (2016). Observed availability of cloud services. master's thesis, University of Helsinki, Faculty
of Science, Department of Computer Science, 1-53.
Palanivel, K., & Kuppuswami, S. (2014). Architecture solutions to E-learning systems using service-oriented
cloud computing reference architecture. Journal of Application or Innovation in Engineering &
Management, 3(3), 547–559.
Pourqasem, J., Karimi, S., & Edalatpanah, S. A. (2014). Comparison of cloud and grid computing. American
Journal of Software Engineering, 2(2), 8–12.
Rad, P., Roopaei, M., & Beebe, N. (2018). AI thinking for cloud education platform with personalized
learning. In Proceedings of the 51st Hawaii international conference on system sciences (pp. 3–12).
Rahadian, R. B., & Budiningsih, C. A. (2017). What are the suitable instructional strategy and media for
student learning styles in middle schools. International Journal on Integrating Technology in Education
(IJITE), 6(4), 25–39.
Riad, B., Ali, S., Mourad, H., & Hamid, S. (2012). An adaptive learning based on ant Colony and
collaborative filtering. Proceedings of the World Congress on Engineering.
Saini, L., Jyoti, & Kaur, H. (2017). Role of cloud computing in education system. International Journal of
Advanced Research in Computer Science, 8, 345–347.
Saleena, B., & Srivatsa, S. K. (2015). Using concept similarity in cross ontology for adaptive e-learning
systems. Computer and Information Sciences, 27, 1–12.
Sasikumar, D., & Saravanakumar, S. (2017). Cloud computing – Research issues, challenges, architecture,
platforms and applications. International Research Journal of Latest Trends in Engineering and
Technology, 5(1), 39–46.
Siddique, A., Durrani, Q. S., & Naqvi, H. A. (2018). Developing adaptive E-learning environment using
cognitive and noncognitive parameters. Journal of Educational Computing Research.
Taniguchi, T., Sakaki, S., Shigenaka, R., Tsuboshita, Y., & Ohkuma, T. (2015). A weighted combination of
text and image classifiers for user gender inference, proceedings of the fourth workshop on vision and
language, 87–93.
Virvou, M., & Troussas, C. (2011). Personalized teaching of multiple languages through the web. Journal for
e-Learning Security (IJeLS), 1, 52–59.
Wainer, J. (2016). Comparison of 14 different families of classification algorithms on 115 binary datasets (pp.
1–36). Machine Learning: Computer Science.
Wexbridge, J. and Nyland, W., (2014). NetBeans Platform for Beginners, This book is for sale at
http://leanpub.com/nbp4beginners.
Wiyono, S., & Abidin, T. (2019). Comparative study of machine learning KNN, SVM, and decision tree
algorithm to predict student's performance. International Journal of Research - Granthaalayah, 7, 190–
196.
Wu C-H, Chen Y-S & Chen TC (2017). An adaptive e-learning system for enhancing learning performance:
Based on dynamic scaffolding theory. International Journal of Mathematics, Science and Technology
Education, 14, 903–913.
Yadav, K. (2014). Role of cloud computing in education. International Journal of Innovative Research in
Computer and Communication Engineering, 2(2), 3108–3112.
Yang, T. C., Hwang, G. J., & Yang, S. J. H. (2013). Development of an adaptive learning system with
multiple perspectives based on students’ learning styles and cognitive styles. Educational Technology and
Society, 16, 185–200.
Zhao, K., Yang, Q., & Ma, X. (2017). Exploration of an open online learning platform based on Google cloud
computing. International Journal of Emerging Technologies in Learning, 12, 17–13.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Sameh Ghallabi ’s student research of Computer Science and Educational Technologies at the University of
Tunis, Tunisia.

Fathi Essalmi is a PhD in computer science since 2011. He has more than 60 journals and conferences papers.
He has several papers published in journals with high impact factor and ranked conferences. Furthermore he
has the habilitation degree to supervise research.
Education and Information Technologies

Mohamed JEMNI is a Professor of Computer Science and Educational Technologies at the University of
Tunis, Tunisia. He is the Director of ICT Department at The Arab League Educational, Cultural and Scientific
Organization (www.alecso.org) from October 2013. He has been the General Director of the Computing
Center El Khawarizmi, the Internet services provider for the sector of higher education and scientific research
in Tunisia, from 2008 to 2013. At ALECSO, he is currently leading several projects related to the promotion
of effective use of ICT in education in the Arab world. He produced two patents and published more than 300
papers in international journals, conferences and books.

Dr. Kinshuk is the Dean of the College of Information at the University of North Texas. Prior to that, he held
the NSERC/CNRL/Xerox/McGraw Hill Research Chair for Adaptivity and Personalization in Informatics,
funded by the Federal government of Canada, Provincial government of Alberta, and by national and
international industries. He was also Full Professor in the School of Computing and Information Systems
and Associate Dean of Faculty of Science and Technology, at Athabasca University, Canada.

Affiliations

Sameh Ghallabi 1 & Fathi Essalmi 1 & Mohamed Jemni 1,2 & Kinshuk 3
1
The Research Laboratory of Technologies of Information and Communication & Electrical engineering
(LaTICE), Tunis, Tunisia
2
ICT Department, The Arab League Educational, Cultural and Scientific Organization, Tunis, Tunisia
3
College of Information, University of North Texas, Denton, TX, USA