Adaptive e-learning is a specialised class of e-learning, in accordance with the philosophy that learners
differ from one another in several ways. For instance, learners differ with respect to their individual
knowledge levels, Learning Styles (LSs), and cognitive abilities. Thus, it is essential for e-learning plat-
forms to adapt and deliver content depending on learner preferences [178, 179]. This is where adaptive
e-learning comes in by allowing the integration of learner characteristics into learner models to deliver
adaptive content. Improving learning efficiency and learner satisfaction rates are the major reasons for
the significant push in the development of adaptive e-learning models [180, 181].
As stated by [182], an adaptive e-learning system constitutes"an interactive system that personalises
and adapts e-learning content, pedagogical models, and interactions between participants in the envi- ronment to meet the individual needs and preferences of users if and when they arise".
Interchangeable terms such as intelligent adaptive/adaptable systems, adaptive educational systems
and personalised learning systems are employed in literature to describe the concept of adaptive e-
learning. Nevertheless, a significant difference exists between an adaptable and an adaptive system where
the former represents a user-initiated adaption technique, while the latter indicates system/automatic-
initiated adaptation techniques without direct user intervention [183]. Generally, adaptation is realised
in either of three forms, namely, adaptive content presentation, adaptive content sequencing as well as
adaptive navigation instruments as shown in Table 2.7 [126, 184].
2.4.1
Components of adaptive e-learning systems
Adaptive e-learning environments are adopted are closely linked to well-organised and structured models.
Figure 2.6 shows the major components of adaptive e-learning systems: student model, a learning objects
Table 2.7: Adaptation categories Adaptation categories Description Adaptive presentation of academic content
Content adapted to tailor to learner preferences constitutes this approach. In addition to ensuring adaptive content to learner characteristic,
another aim of this category involves a substantial increase in the speed as
well as quality of learning [126, 185]. Examples of this category include Arthur [186] and Computer Systems hypermedia courseware (CS383) [117].
Adaptive content sequencing
The purpose of this category involves the order in which the content is presented and delivered to suit various learner preferences, such that the presentation sequence has a significant effect on the learning process [184, 187]. Examples of this category include Adaptive Courseware Environment (ACE) [188] and the (INtelligent System for Personalised
Instruction in a Remote Environment (INSPIRE) [126].
Adaptive navigation tools
Cites the provision of proper orientation of e-learning tools in order to enhance user experience based on students’ habit and preferences by the adjustment of
visible links that control learning orientation such as that of Adaptive Educational System based on Cognitive Styles AES-CS [154].
Figure 2.6: Adaptive e-learning systems’ components
2.4.1.1 Student Model
A student model is responsible for tracking an individual learner’s data in order to adapt itself to the
a crucial piece of individualised behaviour in adaptive e-learning systems and strongly depends on the
way in which knowledge about a student is modelled internally. This process of building and updating
a Student Profile (SP) is known as student modelling, whose phases are described below in Figure 2.7.
Furthermore, student modelling can be classified [5] into static or dynamic modelling as described below:
Figure 2.7: Student modelling in e-learning systems [5]
• Static modelling describes the process of initialising student information once, generally at the time of student registration [191].
• Dynamic modelling is associated with the process of updating student models consistently and respond to changes during the course of enrolment [191].
2.4.1.2 Content Model
Content model generally comprise information related to the knowledge domain of course content in
order to facilitate an adaptive course delivery. The concept of Learning Objects (LO) is made use of to
reduce unnecessary time and effort taken up to develop educational material. [192] explains Learning
Objects to be organised digital materials of learning used in various learning environments and annotated
using metadata for the purpose of describing and manipulating them. As defined by the IEEE [193]. Most
broadly used standards related to metadata in e-learning include those of the Dublin Core [194] and IEEE
2.4.1.3 Adaptive Engine
The adaptive engine is an algorithm that integrates information from the student and course models to
select appropriate course learning objects to present to the student. There are two essential types of
adaptive engines used in for adaptive learning systems [195, 196]:
• Rule-based using conditional ‘if-then’ logical decisions. This approach builds its system as a branching architecture. This approach is clear and simple; however, it becomes more and more
difficult to process the branches as the number of components of the domain model increases
[196–198].
• Algorithm-basedusing mathematical functions to analyse students’ behaviour patterns. This ap- proach is considerably more complex than the Rule Based technique. Algorithm based selec-
tion often uses machine learning techniques to learn more about the contents and the students’
behaviour. These techniques employ highly complex algorithms for predicting which choice of
content is likely to be most successful in terms of learning outcomes [199].
Finally, the recommendation module executes the adaptive matching rules coming from the adaptive
engine and provides the recommendations, as explained in Section 2.5.1.
2.4.2
Adaptive e-learning evaluation methods
In general, the evaluation of adaptive e-learning systems that deploys students’ profiles is considered to be
complicated in addition to being expensive [200]. There is no standard way to evaluate adaptive systems,
but each system might propose different strategies which are tailored based on their purpose and structure.
On the basis of the literature, three main approaches commonly used to evaluate recommendation systems
include offline, user-centred studies and online evaluations [27].
• Existing datasets have been extensively used inoffline evaluationsto measure the performance
of recommender systems through statistical analysis. As will be explained in the next chapters,
we have used students’ datasets from the AAST-MOODLE log-file so as to independently test
the performance of each module of ULEARN system. Within this type of evaluation, there is no
need to deploy the system or interact with real users in an online environment. Another advantage
includes the ability to reuse datasets for testing different algorithms and parameters for the purpose
the way in which it cannot measure users’ opinions of the personalisation experience. However,
user-centred and online approaches are proposed to address some of these limitations.
• Real students are engaged inuser-centred evaluationsfor collating realistic behaviours in addi-
tion to testing the proposed system’s performance [27] and comprises three steps: (i) recruiting
students; (ii) creating a set of tasks to be completed by the users; and (iii) collecting and analysing
student interactions and behaviours. These evaluations are also known to be a better indicator of the
proposed system’s effectiveness, accuracy and efficiency instead of using datasets since systems
are tested in realistic settings. Many studies [201,202,202] confirm that the user-centred evaluation
is suitable for evaluation of the adaptive e-learning system’s accuracy and overall performance.
• A system is made use of inonline evaluationsto be evaluated within a real environment. This
approach is presumed to be the best approach for evaluation purposes since thorough testing of
the system can take place corresponding to real users’ behaviours within a real environment [203].
Therefore, the results obtained through means of an online evaluation would have stronger evi-
dence of the system’s performance. Although online evaluations might be more reliable to assess
a system’s performance, they possess few challenges such as being time-consuming and compli-
cated.