Towards Personalized Content in Massive Open Online Courses

Nour El Mawas

, Jean-Marie Gilliot

, Serge Garlatti

, Reinhardt Euler

and Sylvain Pascual

LabSTICC, IMT Atlantique, Brest, France

LabSTICC, University of Brest, Brest, France

Education Sector, Immanens, Palaiseau, France

s.pascual@immanens.com

Keywords: Personalization, MOOCs, Learner Model, Course Model, Lifelong Learning.

Abstract: Despite the growth of MOOCs, Lifelong learners confront many difficulties related to the attendance of

courses on MOOCs. Lifelong learners are often very different in terms of background, ability, experience,

and prior knowledge but they are required to follow the same content. This explains the low average

completion rate for MOOCs. The research presented in this paper aims to define the functional and technical

architecture to personalize content in Massive Open Online Courses in a Lifelong Learning perspective. The

term content refers to videos, tutorials, documents, exercises, and quizzes in MOOCs. This work is dedicated

to teachers, MOOC designers, MOOC providers, pedagogical engineers, and researchers in e-Learning and

learning analytics. This work takes place within the context of a European project called MOOCTAB

(Massive Online Open Course Tablet).

1 INTRODUCTION

Lifelong Learning (LLL) refers to systematic and

purposeful learning throughout a person’s life

involving formal (schools) and informal (work,

recreation, leisure, social relations, family life)

domains (Cropley 1978). The original concept of

Massive Open Online Courses (MOOCs) is to offer

free and open access courses for a massive number of

learners from anywhere all over the world (Yousef et

al. 2014). Access to and effective use of relevant

information and continuously learning in MOOCs is

essential for lifelong learners. LLL as a concept has

gone through a lot of changes over the years

especially with the arrival of MOOCs and the

increase of their learning resources. The number of

courses (started/scheduled) has grown from about

100 MOOCs in 2012 to almost 4200 starting 2016,

with a duplication of the number of courses between

2015 and 2016. However, according to (Jordan 2014)

by the International Review of Research in Open and

Distributed Learning, the average completion rate for

MOOCs has only been about 6 percent. There is a

growing trend of researches in the possibility of

MOOC personalisation and adaptation in order to

improve users’ engagements, and hence reduce

MOOCs’ drop-out rate problem (Sunar et al. 2015).

In order to understand the reason behind this low

rate, we have relied on the MOOCs annual report

published by the École Polytechnique Fédérale de

Lausanne (EPFL) (“MOOCs Annual Report 2015”

2016) as EPFL is one of the first universities to

experiment with MOOCs, and among the few in

Europe to integrate the use of MOOCs on its own

campus.

The motivation that drives users to register to an

EPFL MOOC varies according to the need of each

learner. Six reasons are behind the registration to the

MOOC: Finding a new job, getting a promotion,

meeting family expectations, earning a higher salary,

solving a specific problem, and helping to pass class.

The “solving a specific problem” motivation is the

main motivation for 60% of the courses. The

academic degrees held by users of the EPFL MOOCs

are very diverse. The highest degree obtained are high

school, associate degree, bachelor degree, master

degree, and doctoral degree. The percentage of

MOOC users who are currently enrolled in an

educational program is low. Only 34% of registered

learners are students (including part-time students).

The remaining enrolees are not in an educational

program. Therefore, it is important to understand that

users do not have the same background.

The diversity of users’ background who followed

El Mawas, N., Gilliot, J., Garlatti, S., Euler, R. and Pascual, S.

Towards Personalized Content in Massive Open Online Courses.

DOI: 10.5220/0006816703310339

In Proceedings of the 10th International Conference on Computer Supported Education (CSEDU 2018), pages 331-339

ISBN: 978-989-758-291-2

331

a MOOC is a key issue (Kizilcec, Piech, and

Schneider 2013). For example, in the matter of the

Analyse Numérique course, 34 % of learners have

Mathematics, Computers, Engineering backgrounds,

21% of learners have Architecture, Civil Engineering

backgrounds, 12% of learners have Education and

Training, 2 % of learners have Business, Finance,

Sales, Management backgrounds, 4% of learners

have Arts, Design, Entertainment backgrounds, 13%

of learners have Construction, Food, Utilities,

Healthcare, Life Sciences backgrounds, and 2% of

learners have Legal, Administration, Social Services

backgrounds. It means that learners do not have the

same prior knowledge for this course.

In this context, the motivation behind our research

work is that (1) differences exist among learners in

terms of background, ability, experience, prior

knowledge, and (2) MOOC platforms unify the

educational content to all learners without taking into

account these differences. According to (Sloep et al.

2011), learners’ personalization and social learning

are essential concepts in Lifelong and Life wide

Learning contexts. The next challenge is about how

to insure adaptive learning that gives each student a

personal experience in a MOOC. (Amo 2013) also

believes that MOOCs should offer student-centered

learning for effective and quality education in order

to meet each individual learner’s learning

expectations in MOOCs. Furthermore, (McLoughlin

2013) and (Knox et al. 2014) point out that MOOCs

environment is convenient for offering personalized

contents and feedbacks to learners based on their

learning goals. This is because MOOCs provides

learning flexibility and sense of independence

between learners and teachers which are important

when implementing personalization in technology

enhanced learning.

This work takes place within the context of the

European MOOCTAB (Massive Online Open Course

Tablet) project. Its main goal is to create a Tablet-

based platform dedicated to LLL (primary,

secondary, higher and continuous) using an on-

demand MOOC platform with a personalized content.

The MOOCTAB project intends to offer a cloud

based European MOOC on Demand platform with a

Plug & Play approach deployable in Europe and

developing countries. This platform is based on

existing technology bricks and existing open source

platforms like edX.

The paper is organized as follows. Section 2

proposes the theoretical background of the study.

Section 3 presents several existing solutions for

personalized MOOCs. Section 4 details our scientific

positioning and defines our functional and technical

solution. Finally, Section 5 summarizes this paper and

presents its perspectives.

2 THEORETICAL

BACKGROUND

In this section, we discuss theoretical background

directly related to the personalized of MOOC content.

Personalization is the process of providing

relevant content based on individual user preferences

or behaviour (Vignette Corp. 2002). It is the explicit

user model that represents user knowledge, goals,

interests, and other features that enable the system to

distinguish among different users (Brusilovsky and

Maybury 2002).

In the e-learning field (U.S. Department of

Education Office of Educational Technology 2010),

personalization is education, where participants have

different learning objectives, depending on their

learning needs. The training is customized, so this is

possible, and personalized instruction may also

provide opportunities for differentiation and

individualization. In this context, differentiation is

education, where participants have the same learning

goals, but the teaching method varies so they adapt to

the individual student's needs. Individualization is

teaching, where the participants also have the same

learning goals, but participants can move forward at

different speeds and relate to a particular content area

or a given activity in different ways, and teaching is

tailored to individual needs.

According to (Germanakos and Mourlas 2006),

personalization is classified in categories: Link

Personalization, Content Personalization, Context

Personalization, Authorized Personalization and

Humanized Personalization. In this paper, we focus

on content personalization. (Ioannidis and Koutrika

2005) defines four forms of content personalization:

information filtering systems, recommender systems,

continuous queries, and personalized searches.

Information filtering systems screen out irrelevant

data from incoming data streams and distribute

relevant data items according to a user profile.

Recommender systems have automated the everyday

procedure of relying on recommendations from other

people whenever personal experience is not sufficient

for making choices. Continuous queries are issued

only once and executed continuously over the

database. Personalized searches are based on the

observation that “to enhance user searches one needs

to take into account the fact that different people find

different things relevant”. In our research work, we

CSEDU 2018 - 10th International Conference on Computer Supported Education

332

are interested in the form of information filtering

systems.

To allow the personalized content, we need to

model the learner. The model must depend on the

learner himself and the domain which is the course in

our case. The next section details existing projects on

MOOC personalization. Note that we consider the

personalization as a specific concept of the adaptation

where adaptation is based on the personal preferences

and background of the learner.

3 RELATED WORK

In this section, we consider existing projects related

to personalized MOOCs and we deduce important

elements to ensure this personalization.

3.1 The MOOC Personalization for

Various Learning Goals project

The MOOC Personalization for Various Learning

Goals project is a project funded by the Bill and

Melinda Gates foundation. It aims to identify how

students’ goals are expressed through their activities

on the edX learning platform, and how they evolve

over time.

The objectives of this project were: 1) classify

student learners by learning goals; 2) cluster learners

by engagement with the platform, comparing various

groups by learning outcomes (i.e., certificate

attainment), and aiming to predict user transition

from one cluster to another; 3) study how the

clustering could be used for platform customization

and personalization of learning experience.

This research was expected to proceed in the

context of HarvardX, (Harvard’s division for online

learning) and to be based on the data on 17 HarvardX

courses running on the edX platform, focusing on 5

courses that must be completed by December 2013.

Since December 2013, there are no research papers

that concern the project.

3.2 The POEM Project

The POEM (Personalised Open Education for the

Masses) project aims at designing a platform that

reconciles Massive Education — as with the strong

development of MOOCs (Massive Open Online

Courses) — with Personalized Education. According

to (Collet 2013), one of the important concepts that

allows personalized education is the deconstruction

of courses and curricula into hundreds and thousands

of short independent units that will interact together

as a complex system. The objective is then to get

these thousands of small independent courses to self-

organize into optimal pedagogical paths that allow

individual students to validate curricula as fast as

possible depending on their personal skills, aims and

previous knowledge. POEM is developed under

Creative Commons and will be as interoperable with

edX. Students involve in many individual and

collective educational activities for their mutual

benefit: assessment, inter-tutorship and construction

of dynamical Knowledge Maps of domains to provide

different learning paths to learners.

3.3 The Knowledge Map on Khan

Academy

Khan Academy proposes math courses with a

knowledge map that makes learning objectives and

individual progress available to learners. The

motivation behind the map is that learners miss an

overview of how all the math exercises tie in together.

The concept of the Knowledge Map is behind the

Math Missions in the sense that exercises build on

another and basic concepts are introduced before

advanced ones. This knowledge map is in forms of

skill-meter (display and badges) (Thompson 2011). It

contains a starry night, containing all of the stars. The

stars represent lessons. Yellow stars with a blue

border are lessons, users are proficient at, green

borders mean recommended lessons, and others are

lessons that are not recommended. An orange border

means a lesson a user should review. It also tells the

user how skills are connected to each other. The

Knowledge Map also has a navigation bar, with

which students could search for a particular skill.

3.4 The ECO Project

Brouns et al. (2014) proposes the European ECO

(Elearning, Communication and Open-data: Massive

Mobile, Ubiquitous and Open Learning). The

motivation behind this project is that MOOCs are

proving to be inconsistent with the European

standards for formal higher education due to their

low-level of learner support and lack of an enriched

pedagogical approach. This project introduces the

notion of sMOOCs (“social” MOOCs) which

provides a learning experience marked by social

interactions and participation.

The sMOOCs are accessible from different

platforms and through mobile devices and integrated

with participants' real life experiences through

contextualization of content via mobile apps and

gamifications. It also supports adaptive learning

Towards Personalized Content in Massive Open Online Courses

333

strategies and ubiquitous, pervasive and

contextualized learning. ECO sMOOCs have the

potential to adapt to the changing intentions of

participants during the course.

3.5 The aMOOC Project

(Sonwalkar 2013) proposes an adaptive MOOC

(aMOOC) platform, providing a strong pedagogical

framework and a personalized learning experience in

a MOOC learning environment. The aMOOC allows

for different ways to organize content, offering

different context and perspective for learners. It also

aims to identify the way a learner would like to learn

by conducting diagnostic assessments on the learning

preference. It uses assessment results to provide

continuous intelligent feedback that motivates and

provides guidance to overcome concept deficiencies

and maximize learning performance.

In this project, learning strategies are related to

five learning pedagogies: apprentice (learning

through mentor–student interaction), incidental

(learning through case study), inductive (learning

through example), deductive (learning through

application), and discovery (learning through

experimentation). The content of the aMOOC is

presented to students based on the learning style of

preference. For example, in the incidental learning

study, learning happens primarily within a context of

case studies. Content provided by the expert is

sequenced in ways that explain the events involved in

the case study.

3.6 Discussion

This state-of-the-art allows us to define important

elements for our content personalization approach

(Table 1): learning goals, learning experience,

learning recognition, learning path, and content

granularity.

Note that for clarity reasons, in Table 1, E1 refers to

learning goals, E2 to learning experience, E3 to

learning recognition, E4 to learning path, E5 to

content granularity, P1 refers to the MOOC

Personalization for Various Learning Goals project,

P2 to the POEM project, P3 to the knowledge map on

Khan Academy, P4 to the ECO project, and P5 to the

aMOOC project.

The learning goals are a key element in content

personalization. It is a very personal decision that has

its roots in a social environment providing examples,

discussions and opportunities. A learner has a set of

realistic and achievable goals and based on these

goals the content must be delivered to him. The

learning experience refers to Learning by doing

which takes place through on-the-job and leadership

experiences. The learning recognition is important in

our approach. It acknowledges achievements and

constitutes certified evidence. It includes formal

learning such as diplomas, certificates, and

recommendations. The learning path makes learning

objectives and individual progress available to

learners. It allows an overview of how all learning

concepts tie in together and where is the learner's

current position in the learning path. The content

granularity is related to the pieces of learning content

that are combined to form the whole MOOC content.

For example, if a content package is comprised of

only a few pieces of large grained learning content

then re-sequencing them to form a new learning path

for another learner may not be possible. This issue is

paramount in the delivery of any personalized

content.

Table 1: Important elements / levels for content

personalisation based on existing projects.

Learning

Visualisation

Content



These elements can be categorized in three levels

(Table 1): (1) the learning level includes learning

goals, learning experience, and learning recognition;

(2) the visualization level includes the learning path;

(3) the content level includes the content granularity.

To highlight all these ideas, we are going to detail

in the next section our approach that takes into

account these elements and provides innovative

solutions in this domain.

4 OUR PROPOSED APPROACH

In this section, we present an overview of our

approach. Then we detail our functional architecture

and our Domain / Learner Models before discussing

the presence of our elements categorized in three

levels as defined in Section 3.6.

4.1 An Overview of Our Approach

The difference between a course completion in a

classic MOOC and in our approach is the

personalization of the course content.

CSEDU 2018 - 10th International Conference on Computer Supported Education

334

Figure 1: The course completion.

Figure 1 shows how the personalization occurs

during the course completion. The learner logins in

the MOOC platform. He can, therefore, choose a

course to take. Before starting the course, the

platform asks him to fulfil a positioning

questionnaire. This questionnaire is about the current

professional situation, his diplomas, his certifications,

and the platform permission to access to his LinkedIn

profile. Once the questionnaire is submitted by the

learner, the platform analyses the questionnaire

response and creates the Learner Model for the

learner.

Note that the Learner Model is addressed in

Section 4.2. Based on the Learner Model and while

the course is not completed, the platform proposes a

personalized content to each learner who can interact

with it. Then the learner will be evaluated on this

specific content before updating his Learner Model.

In the next section, we will detail our functional

architecture that allows this personalization.

4.2 Our Functional Architecture

Our learning architecture (Figure 2) is designed in

order to be compliant with different MOOC platform

architectures. In general, MOOC platforms

distinguish two main components dedicated to

different steps in the course lifecycle: the Content

Management System (CMS) and the Learning

Management System (LMS). The CMS is used to

manage students’ enrolment, track students’

performance, and create/distribute course content.

The LMS focuses on course management including

user registration, tracking courses, recording data

from learners, and analysis purposes.

In our vision, we consider three main roles: the

pedagogical engineer, the teacher, and the learner. In

a standard course creation, the pedagogical engineer

has to provide the course structure and populate it

with the course content. In our approach, the course

structure is becoming a part of the Domain Model

(DM). We propose an LMAP editor that enables to

define the structure of the Domain Model with related

content and provision of potential exercises. The

LMAP editor replaces the classical linear description

of a course in traditional platforms while the content

description does not change. When the DM is created,

the course structure and content are uploaded by the

pedagogical engineer in the LMS.

When the learner will access the course, he will

get personalized content through our “Course

Navigation” plug-in. Content will be proposed

according to his own current Learner Model (LM). He

can also visualize his current progress through the

LM Dashboard and point specific topics in the DM.

Other MOOC activities such as forums and quizzes

are maintained in our approach.

Teachers have standard access to learner progress

and productions on the platform. They have also

aggregated access to LM of the learners registered in

their course.

Figure 2: Our functional architecture.

Now we will detail the domain and the Learner

Models which are main elements in our approach.

Towards Personalized Content in Massive Open Online Courses

335

4.3 Domain and Learner Models

Our Domain Model is shown in Figure 3. It has three

layers: subject, topic, and concept. The Domain

Model is composed of a set of subjects, each subject

is composed of many topics, and each topic refers to

many concepts.

Figure 3: The structure of our Domain Model.

Our Learner Model (Figure 3) is based on the

Generic Bayesian Student Model (GBSM) (Millán et

al. 2013). It is composed of two different kinds of

variables: knowledge and evidential variables.

Knowledge variables (K) represent students’

knowledge (either declarative or procedural

knowledge, but also skills, abilities, etc). These are

the variables of interest in adaptive e-learning

systems, in order to be able to adapt instruction to

each individual student. Their values are not directly

observable (i.e., they are hidden variables). In the

GBSM, all knowledge variables are modelled as

binary, and take two values: 0 (not-known) and 1

(known).

Evidential variables (Q), which represent

students’ actions, are directly observable. For

example, the results of a test, question, problem

solving procedure, etc. The values of such variables

will be used to infer the values of the hidden

knowledge variables. In the GBSM, evidential

variables are also considered to be binary, with values

0 (incorrect) or 1 (correct).

Figure 4: The structure of our Learner Model.

In Figure 4, there are two types of relationships:

aggregation relationships and causal relationships.

Aggregation relationships are between knowledge

nodes (basic concepts, topics and subject). Causal

relationships are between knowledge and evidential

nodes (concepts and evaluations).

4.4 Our Technical Architecture

Technically, our architecture (Figure 5) relies on

three main components: the learner environment, the

Learning Record Store (LRS), and the Learning Map

(LMAP) core.

The learner environment is composed of different

learning tools. The LMS platform is the main

component of this environment. It contains the

Course Navigation module that gives the learner a

personalized access to content. In the learner

environment, MOOCs are central but there are also

other assessment platforms and social networks

offering learning services.

Since we have different learning services and

platforms, we need to collect learning experience and

performance data from many different sources and

present them in a meaningful way. That is why we

choose the use of the LRS that supports the open

standard, xAPI (Experience Application Performing

Interface). In this way, all learning traces collected

from the learner environment are transferred to the

LMAP core via the LRS. Note that a statement (to be

approved by the teacher) can be made by the user

himself based on a certification or on a

previous/current job.

The Learner Models are dynamic and must be

updated. As such, we used the LMAP core to (1) store

the Domain and the Learner Model, and (2) update

the Learner Models. In the LMAP core, we have two

main components and two interfaces. The main

components are the Learner Model Updater (LMU)

and the Selector. The LMU updates the Learner

Model based on new assessments and learner

achievements collected by the LRS. The Selector

chooses the personalized content from the Domain

Model according to the current Learner Model. The

access to the models is provided separately by the

Domain Model (DM) Interface and the Learner

Model (LM) Interface. The DM interface enables

Domain Models creation, modification, and deletion.

It is defined for the DM editor in the CMS. The LM

interface enables achievement updates, and access. It

enables interactions with the learner and the teacher

through LM Dashboard in the LMS.

Our first implementation is based on the edX

platform, as it is the main open source platform with

an active developers’ community. We have

developed xAPI connectors in order to collect learner

traces of statements. Course Navigation is integrated

CSEDU 2018 - 10th International Conference on Computer Supported Education

336

by using LTI standard that permits seamless

integration of external components.

Figure 5: Our technical architecture.

As we explain in Section 4.3, the pedagogical

engineer defines the Domain Model. The Domain

Model is created via the LMAP editor which we have

developed for this purpose. The frontend of our

LMAP editor is based on Javascript, html, css, and

svg. The backend is created using open source

software LAMP (Linux-Apache-Mysql-PHP) server

technology and PHP-framework Symfony 2. When

the pedagogical engineer adds a new element

(subject, topic, concept, or evaluation) in the LMAP

editor, he needs to define properties below: the name

of the element (label), its priority, the order it has in

relation to other elements, its acquisition link (link to

an online content), its acquisition mode, its validation

link (if it exists), its validation approval, and the

number of hours and weeks for acquisition.

4.5 Discussion

Our functional and technical architectures take into

account the important elements for MOOC content

personalization as detailed in Section 3.6 (see Table

2).

At the learning level, the positioning

questionnaire (Section 4.1), the statements made by

the user himself based on a certification or based on a

previous/current job, and all learning traces are

transferred to the LMAP.

At the visualization level, the LMAP shows the

learning path of the learning and his current position

in the learning path.

At the content level, we have three layers of

granularity: subject, topic, and concept (Section 4.3).

These layers are comprised of a large number of

pieces of small grained learning content which allow

to re-sequence them to form personalized learning

paths for each learner.

Table 2: The presence of the important elements / levels for

content personalization in our approach.

Learning

Visualisation

Content

Statements

+ traces



LMAP



3 layers of

granularity



To summarize, in this research work, we propose

a functional and a technical architecture to allow

personalized content for each learner who attends a

MOOC course.

5 CONCLUSION AND

PERSPECTIVES

This study addresses the problem of unified content

in Massive Open Online Courses for Lifelong

learners. The main questions of the study are how to

address differences between learners (in terms of

background, ability, experience, prior knowledge),

what are the approaches allowing MOOCs to take

into account these differences, and how to promote

personalized content in MOOCs in order to propose

suitable content and increase learning among

learners.

We investigate the problem from its theoretical

background, and we consider existing approaches

related to personalized MOOCs in order to see if any

existing approach can meet our requirements.

Unfortunately, no one can respond to our needs in

terms of the support of learner's level of knowledge,

learner’s background, learning goals, navigation

preference, and the presence of a concept map for the

course and a graphic path indicator. To achieve this,

our approach is proposed as a functional and technical

solution to our problem. This solution allows

personalized content in MOOCs. Thanks to this

solution, learners in MOOCs have more choice; they

take more ownership of their learning and develop

their learning strategies as well as self-regulated

Towards Personalized Content in Massive Open Online Courses

337

learning behaviours that are necessary for meeting

immediate goals and for LLL.

Now we will refine our learner and domain

models and implement them before deploying our

solution in classrooms in France and Turkey, in the

framework of the MOOCTAB project. Then we will

evaluate our approach, focusing particularly on

results achieved in terms of knowledge learning by

learners. For that purpose, the learning will be

estimated by placing the learners in two groups: for a

controlled period of time, the first group will attend a

course on a standard MOOC platform and the second

group will attend the same course on our personalized

MOOC platform. The selection of the learners is

based on a preliminary questionnaire to test

prerequisites for each learner and to drive down

inequalities in knowledge. The content of this

questionnaire also depends on the knowledge

addressed in the course which confronts learners in

order to decrease knowledge heterogeneity of the two

groups. To interpret the evaluation results, we will

base on different variables tracked by our platform

and that we consider as learners’ traces such as

learning outcomes (i.e., course completion, course

grades) and parameters related to the platform use

(time spent on watching videos, on answering

questions, on passing an exam). These variables will

be used to compare the various learners in the two

groups. Next, we will consider how learners’

interactions with the platform evolve over time in

order to track changes in their learning goals.

ACKNOWLEDGEMENTS

This research was supported by The ITEA 2

(Information Technology for European

Advancement) Massive Online Open Course Tablet,

MOOCTAB (2014-2017) project.

REFERENCES

Amo, Daniel. 2013. “MOOCs: Experimental Approaches

for Quality in Pedagogical and Design Fundamentals.”

In Proceedings of the First International Conference on

Technological Ecosystem for Enhancing

Multiculturality, 219–223. ACM.

Brouns, Francis, José Mota, Lina Morgado, Darco Jansen,

Santiago Fano, Alejandro Silva, and António Teixeira.

2014. “A Networked Learning Framework for Effective

MOOC Design: The ECO Project Approach.”

Brusilovsky, Peter, and Mark T Maybury. 2002. “From

Adaptive Hypermedia to the Adaptive Web.”

Communications of the ACM 45 (5): 30–33.

Collet. 2013. “POEM (Personalised Open Education for the

Masses).” Educpros : Actualités et Services Pour Les

Professionnels de l’éducation. 2013.

http://www.letudiant.fr/educpros/.

Cropley, Arthur J. 1978. “Some Guidelines for the Reform

of School Curricula in the Perspective of Lifelong

Education.” International Review of Education 24 (1):

21–33.

Germanakos, Panagiotis, and Constantinos Mourlas. 2006.

“Adaptation and Personalization of Web-Based

Multimedia Content.” Digital Multimedia Perception

and Design, 284–304.

Ioannidis, Yannis, and Georgia Koutrika. 2005.

“Personalized Systems: Models and Methods from an

Ir and Db Perspective.” In Proceedings of the 31st

International Conference on Very Large Data Bases,

1365–1365. VLDB Endowment.

Jordan, Katy. 2014. “Initial Trends in Enrolment and

Completion of Massive Open Online Courses.” The

International Review of Research in Open and

Distributed Learning 15 (1).

Kizilcec, René F, Chris Piech, and Emily Schneider. 2013.

“Deconstructing Disengagement: Analyzing Learner

Subpopulations in Massive Open Online Courses.” In

Proceedings of the Third International Conference on

Learning Analytics and Knowledge, 170–179. ACM.

Knox, Jeremy, Jen Ross, Christine Sinclair, Hamish

Macleod, and Siân Bayne. 2014. “Mooc Feedback:

Pleasing All the People.” Invasion of the MOOCs 98.

McLoughlin, Catherine Elizabeth. 2013. “The Pedagogy of

Personalised Learning: Exemplars, MOOCS and

Related Learning Theories.” In EdMedia: World

Conference on Educational Media and Technology,

266–270. Association for the Advancement of

Computing in Education (AACE).

Millán, Eva, Luis DescalçO, Gladys Castillo, Paula

Oliveira, and Sandra Diogo. 2013. “Using Bayesian

Networks to Improve Knowledge Assessment.”

Computers & Education 60 (1): 436–447.

“MOOCs Annual Report 2015.” 2016. MOOCs Annual

Report 2015. Center for Digital Education , Ecole

Polytechnique Fédérale de Lausanne.

Sloep, Peter, Jo Boon, Bernard Cornu, Michael Klebl, Paul

Lefrere, Ambjörn Naeve, Peter Scott, and Luis Tinoca.

2011. “A European Research Agenda for Lifelong

Learning.” International Journal of Technology

Enhanced Learning 3 (2): 204–228.

Sonwalkar, Nishikant. 2013. “The First Adaptive MOOC:

A Case Study on Pedagogy Framework and Scalable

Cloud Architecture—Part I.” In MOOCs Forum, 1:22–

29. Mary Ann Liebert, Inc. 140 Huguenot Street, 3rd

Floor New Rochelle, NY 10801 USA.

Sunar, Ayse Saliha, Nor Aniza Abdullah, Su White, and

Hugh C Davis. 2015. “Personalisation of MOOCs: The

State of the Art.”

Thompson, Clive. 2011. “How Khan Academy Is Changing

the Rules of Education.” Wired Magazine 126: 1–5.

U.S. Department of Education Office of Educational

Technology. 2010. “Transforming American Education

CSEDU 2018 - 10th International Conference on Computer Supported Education

338

- Learning Powered by Technology. National

Education Technology Plan.”

Vignette Corp. 2002. “Personalization Strategies-Fit

Technology to Business White Paper.,” 2002, TX:

Author edition.

Yousef, Ahmed Mohamed Fahmy, Mohamed Amine

Chatti, Ulrik Schroeder, M Wosnitza, and H Jakobs.

2014. “A Review of the State-of-the-Art.” Proceedings

of CSEDU, 9–20.

Towards Personalized Content in Massive Open Online Courses

339