ADOPT: A Trace based Adaptive System

Henda Belaïd Ajroud

, Insaf Tnazefti-Kerkeni

and Bénédicte Talon

LIPAH, Faculté des Sciences de Tunis, Le Belvédère, Tunisie

LISIC, Université du Littoral Côte d’Opale, Calais, France

Keywords: Adaptive Learning, Traces, Learner Model, Domain Model, Adaptation Model, Agents.

Abstract: Adaptive learning is learning that tries to get closer to the learner in order to identify his/her strengths and

weaknesses and to provide him/her with a learning that best adapts to his/her needs, thereby increasing

his/her chances of success. It is in this spirit that this work was carried out. It is interested in adaptation of

learning when using a Learning Management System (LMS). To achieve our goal, we designed different

models such as the Learner Model and a multi-agents system, ADOPT, which defines intelligent interactive

agents. These agents analyze the traces left by the learner, calculate various indicators and propose the most

suitable adaptations for the learner.

1 INTRODUCTION

During the past decade, the use of e-learning

platforms to supplement or replace face-to-face

training has increased. In some situations, the

teaching is even 100% distance. Unlike traditional

learning where instruction is directed by the teacher

and where memorization is very important, modern

learning is learner-centered and performance-

oriented (Chachoua, 2019). So, some learning

systems or platforms consider the learners’ needs

and adapt to these needs. With these adaptive

learning systems, each learner characteristics are

considered: strengths, weaknesses, specific learning

rhythm and more generally, profile and goal. These

parameters are used to in order to maximize their

learning outcomes and minimize the risks of

abounding. Thus, adaptive learning systems identify

the needs and interests of each learner to provide

personalized contents and specific learning paths.

Some adaptive learning systems have been set up

for specific courses. For example, Yixue is used in

China for mathematics courses among middle school

students (Wang et al., 2019) and in an after-school

english language arts course (Li et al., 2019) and

BAGHERA (Pesty and Weber, 2004) is used for

geometry proof learning. Other systems, such as

AMAS (Gaffney et al., 2014), help users to create

adaptive e-learning components but are not open and

flexible enough to be used in different learning

fields. Other systems adapt presentation such

Allegro (Viccari et al., 2007) which supports

collaborative learning and MASPLANG (Peña et al.,

2002) which supports distance learning via the web.

Therefore, research in the field of adaptive

learning systems is still on the agenda.

This paper is organized in six sections. Section 2

presents a literature review. Section 3 presents

ADOPT (Adaptation Done On-line through a

Process controlled by Tracking), the adaptive e-

learning system we propose and highlights its

models. Section 4 presents the system architecture

and the principle agents used. Section 5 makes a

brief comparison of ADOPT with some existing

adaptive systems and section 6 concludes the paper

and presents our future work.

2 LITERATURE REVIEW

There are three common adaptive e-learning

approaches (Ennouamani and Mahani, 2017; Apoki

et al., 2019).

The first is a macro-adaptive approach. It takes

into account the differences between learners and

adapts the learning rate accordingly.

The second is an aptitude-treatment interaction

approach. It identifies the learner’s main skills by

analyzing the behavior. Different levels of control

occur during the learning process.

The third approach is a micro-adaptive approach.

The learner model evolves according to the learner's

Ajroud, H., Tnazefti-Kerkeni, I. and Talon, B.

ADOPT: A Trace based Adaptive System.

DOI: 10.5220/0010452702330239

In Proceedings of the 13th International Conference on Computer Supported Education (CSEDU 2021) - Volume 2, pages 233-239

ISBN: 978-989-758-502-9

233

interactions with the system. Identifying the needs

allows offering the most appropriate learning process.

The architecture of an adaptive e-learning system

is based on three models: the learner model (also

named student or user model), the domain model

(also named content or expert model) and the learning

model (also referred to as tutoring or pedagogical or

adaptation model) (Vandewaetere et al., 2011).

The learner model is the pivot of adaptive

systems (Kaya and Altun, 2011; M. A. Tadlaoui et

al., 2016; Yue Gong, 2014; Ezzat Labib et al., 2017).

It contains static and dynamic data related to the

learner such as personal properties and behavior

when using the e-learning system (Wartiningsih and

Surjono, 2019). It is as useful for tutors and learners.

Indeed, the tutors can correctly and precisely

evaluate learner capabilities and even predict a

probable success (or not) during learning activities.

Learners can identify nature of the problems they

encountered during the learning process, can

evaluate skills, etc.

Standards of learner model have been defined:

The IEEE standard PAPI learner (Public and Private

Information for Learners) (Farance, 2000), IMS-LIP

(IMS Learner Information Package) (IMS, 2001)

and IMS RDCEO (IMS Reusable Definition of

Competency or Educational Objective) (IMS, 2002).

(Hlioui et al., 2016) explore these standards,

considering parameters such as personal

information, preferences, competencies, etc.

According to these authors, each standard presents

shortcomings. To face these shortcomings, some

researches like (Madhour et al., 2006), (Lazarinis et

al., 2009) and (Ghallabi et al., 2015), use a

combination of IMS-LIP and PAPI standards. But as

noted in (Kaya and Altun, 2011), these standards are

so detailed that they are complex to use. Moreover,

the needs of users can not always be satisfied by

these standards. So, some authors proposed specific

learner models such as in (El-Kechaïi et al., 2015),

(Tack et al., 2016), (Mediani et al., 2015), (Tmimi et

al., 2017) and (Heng et al., 2018).

To collect necessary data and build the learner’s

model in the LMS, it is possible to use

questionnaires and quizzes. Felder and Soloman

propose a questionnaire that can be filled by the

learner in order to identify the learning style

(Radwan, 2014). Exploration of traces left by the

learner during activities on the LMS is another

interesting source of data.



Available at: https://www.webtools.ncsu.edu/learning

styles/



The domain model “is a representation of the

essential learning content present in the system”

(Apoki et al., 2019). It stores expert knowledge and

pedagogical digital resources. It contains all

information about the courses (Ahmed et al., 2017).

Thanks to the information stored in the learner

model, adaptations can be proposed. Indeed, the

learning model presents the rules of adaptation and

describes instructional strategies and the pedagogical

theories (Apoki et al., 2019).

3 PROPOSED ADAPTIVE

e-LEARNING SYSTEM

ADOPT follows the aptitude-treatment interaction

approach. We track learners’ activities during the

learning process on the LMS (Tnazefti-Kerkeni et

al., 2020; Talon et al., 2013). It uses intelligent

agents. In this section, we represent the knowledge

that is the basis of our system. This knowledge is

contained in models inspired by those conventionally

used in an Intelligent Tutoring System.

3.1 Domain Model

Figure 1: Domain model.

Course

Name

Teacher

Syllabus

…

Scenario

Name

Objective

Duration

…

1..*

Resource

Name

Type

Level

…

Test

Name

Duration

Level

…

3..*

Step

Name

Objective

Duration

Level test

…

1..*

Activity

Name

Objective

Duration

…

0..*

CSEDU 2021 - 13th International Conference on Computer Supported Education

234

The domain model describes learning objectives,

courses, syllabus as well as learning scenarios. The

digital learning resources can be text, video, audio,

etc. The tutor presents a course as a set of scenarios.

Each scenario S is a list of steps given in a certain

order.

Each step has an objective and duration and

presents three levels: beginner, intermediate and

advanced. At the beginner level, only the basic

concepts are presented. The intermediate level is

more detailed than the first one and the advanced

level is an expert level. For each level, the tutor

provides visual, auditory, and kinesthetic oriented

resources. The tutor also provides tests for each

level. A test can be a QCM or an exercise and can

eventually plan activities. Figure 1 presents an UML

class diagram of the domain model.

We consider that a step is achieved by a learner

if at least 90% of the tests are done successfully. A

scenario S is acquired if all its steps are done.

3.2 Learner Model

The learner model establishes the profile of the

learner by providing his characteristics and the

activities done.

When a learner registers on the LMS, he/she

enters data such as name, age, gender, and other

static data. He/she passes a test enabling to

determine his/her personality data. Felder and

Soloman questionnaire and Felder and Silverman

learning style model (Felder and Silverman, 1988)

inspired our test.

Figure 2: Learner Model.

In addition, each action done by the learner in the

LMS is tracked. These traces are analyzed, modeled,

and stored. They are used to make real time updates

of the learner model.

The learner model used in the proposed adaptive

e-learning system is represented in figure 2 as an

UML class diagram. It regroups information about

the learner. Indeed, we have:

 Learner personal data,

 Pedagogical data which consider the

knowledge of the learner in each course and

which evolve with learning.

 Personality data which consider his

learning style (visual, auditory, kinesthetic)

and personality type (perception: sensing or

intuitive, processing: active or reflective

and understanding: sequential or global).

3.3 Adaptation Model

The adaptation or learning model proposes

adaptations according to the learner profiles. The

rules of this model are defined using indicators.

When a learner registers in a course, he/she

begins with a pre-assessment which identifies the

knowledge he/she has mastered in this course

according to the course knowledge map (hierarchy

which is defined in the domain model). In this way,

the system identifies the starting position of this

learner for the course and each learner will start the

course from his knowledge level. From his starting

point and for each step of a scenario, he/she must

start by taking a level test. If he/she gets under 50,

than he/she has a beginner level in this step. If

he/she gets between 50 and 70, then his/her level is

intermediate, otherwise it is advanced. The learner

gets resources and tests corresponding to the level

and when he/she finishes a level, he/she goes to the

next level and so on until he/she finishes the step.

3.3.1 Indicators

We note S={s

, s

, …, s

} such as s

is the i

step of

the scenario S. We note respectively R

and T

, the

resource and the k

test presented in the step s

Thus, if l>k than T

should be more difficult than T

and it is proposed to the learner only if T

is well

done.

We consider R

as the minimum time required

by the learner to complete the study of the resource

and T

as the maximum time required by the

learner to do the test T

Here are some of the indicators used by the

system.

Learner

data

Identity

Gender

Age

…

Pedagogical

data

Course

Scenario

Step

Level

…

Personality

data

Learning

style

Perception

Processing

Inderstanding

…

1..* 1

ADOPT: A Trace based Adaptive System

235

 I1

: Reading or viewing or listening time used

by the learner to study the resource R

 I2

: Duration of the test T

done by the learner.

 I3

: Number of times the test T

is repeated

by the learner before it is well done.

3.3.2 Rules

We present below some of the rules of the

adaptation model:

 Rule 1: It is a rule to adapt the resource to the

learner style. If the learner style is visual, then

the use of video, PowerPoint, picture, etc.

resources for the step is proposed. If,

contrariwise, it is auditory, then podcast or

video resources, etc. are proposed.

 Rule 2: If the learner spends a little time

studying a resource (I1

<R

) then the system

suggests him to review again the same

resource and/or to study other resources from

the same level.

 Rule 3: If the learner spends too much time to

well do a test (I2

>T

and T

is well done)

then the learner is not comfortable with this

test and step and the system proposes him to

review a resource of this step and to do

another test with the same difficulty before

continuing.

 Rule 4: If the learner spends too much time to

do a test without success or if he/she repeats

the test several times before doing it correctly

((I2

>T

and T

is not well done) or I3

2)

than the learner has difficulties with this step

and the system proposes to review resources

of this step before doing another test with the

same difficulty.

 Rule 5: If, after three attempts, the learner still

fails to well do the test, he/she is then given its

detailed solution with explanations and the

system suggests him to review the course

resources relates to this part and proposes

another test of the same level of difficulty.

 Rule 6: If the learner does the test correctly in

a very short time (I2

<<T

), then the system

suggests that he/she goes directly to the last

test of this step.

 Rule 7: If, after a certain number of tries doing

a test the learner always fails (T

is always not

well done) then the system suggests to review

the resource and to do the lower level test

i(k-1)

4 SYSTEM ARCHITECTURE

ADOPT is based on traces. Any exchange or

interaction between the learner and the system is

noted, modeled and stored in a trace model.

ADOPT is a Multi-Agent System. Each model

presented above is based on an interactive agent.

When interacting, agents update their knowledge and

adapt their behavior. They provide the knowledge

used to carry out the pedagogical reasoning.

Here is a list of the main agents involved in the

system as well as a brief presentation of their main

role:

 a-Observer agent tracks the learner's actions

when using the LMS. It cleans the raw traces

Figure 3: System architecture.

LMS

istration

Use

Learner

Tutor

Trace model

Learner

model

Domain

model

Adaptation

model

a-Observe

a-Learne

a-AssistL

a-Ex

ert

Adaptation proposal

a-AssistT

Indicators

Resources, scenarios, tests,

quiz, MCQ, etc.

CSEDU 2021 - 13th International Conference on Computer Supported Education

236



Figure 4: Example of the system interface.

and treats them to get the modeled traces and

stores them in the trace model.

 a-Learner agent defines the learner model and

updates it in real time to consider the

evolution of the learner in his learning.

 a-Expert agent is based on the rules defined in

the adaptation model and on the knowledge

stored in the domain model and in the learner

model to propose adaptations to the learner. It

recognizes the knowledge level of the learner

and updates the learner model accordingly.

 a-AssistL agent presents some indicators on

the dashboard of the learner which help the

learner to see his/her difficulties and to get

feedbacks regarding his/her learning.

 a-AssistT agent presents to the tutor tables and

graphs which help to monitor learners’

learning.

Figure 3 presents the system architecture and

figure 4 shows an example of the system interface.

5 COMPARISON WITH OTHER

ADAPTIVE SYSTEMS

We compare here ADOPT with other existing

adaptive systems using agents.

In addition to other criteria, we consider the

criteria defined by Brusilovsky (Brusilovsky, 1996)

for adaptive hypermedia and which are listed below:

 Direct guidance: Suggesting to the learner

what to do next.

 Adaptive sorting: Sorting all the links

according to the learner model.

 Adaptive hiding: Hiding links according to the

learner model.

 Adaptive annotation: Giving a comment or

annotation on the state of a link (for example

visited link or not yet visited link).

 Adaptive presentation: adapting presentation

to knowledge level and other characteristics of

the learner such as his learning style.

As we notice it in table 1, even if ALLEGRO

and MASPLANG are not domain dependent, they

are only interested by adaptive presentation.

Course title

% of target

knowled

e ac

uire

% of target course

acquired

Target knowledge :

Step title

Level Test

Link to

resources

Explanation

requested

Send email to

the tutor

Explanation

ADOPT: A Trace based Adaptive System

237

Table 1: Comparison with adaptive Multi-Agents Systems.

System

Criterion

BAGHERA ALLEGRO MASPLANG ADOPT

Direct guidance - - - X

Adaptive sorting - - - X

Ada

tive hidin

X --X

Ada

tive annotation X - - X

Ada

tive

resentation X X X X

Domain dependent X - - -

Learner characteristics :

 Knowledge level

 Learning style

 Learnin

ress

6 CONCLUSION AND FUTURE

WORK

Adaptive learning systems are a family of learning

systems that interests in learners’ needs and profile

learning to increase success. In these systems, the

characteristics of each learner are considered and

this in order to consolidate the acquisition of

learning and to minimize the risk of dropping out

and failure. We interest in adaptive learning systems

that identify needs and interests of learners to

provide personalized contents and specific learning

paths. In this paper, after a literature review, we

have described the different models integrated in our

system and the agent architecture designed to

support tracking and personalization. A final

comparison of ADOPT (Adaptation Done On-line

through a Process controlled by Tracking) with other

ones has highlighted innovations it offers. The

global architecture allows an in real-time adaptation

management. Some features are not completely

developed. So the future of our work will consist in

a total implementation of functionalities. Then we

will be able to test and evaluate ADOPT in a real

higher dimensional context.

REFERENCES

Ahmed, M. U., Sangi, N. A. and Mahmood, A., 2017. A

Learner model for adaptable e-learning, International

Journal of Advanced Computer Science and

Applications, Vol. 8, No. 6, 2017.

Apoki, U. C., Al-Chalabi, H. K. M. and Crisan, G. C.,

2019.From Digital Learning Resources to Adaptive

Learning Objects: An Overview, In the proceeding of

the 6th International Conference, MDIS 2019,

Romania, October, 2019.

Brusilovsky, P., 1996. Methods and techniques of adaptive

hypermedia, In User Modeling and User Adapted

Interaction, 1996, v 6, n 2-3, pp 87-129(Special issue

on adaptive hypertext and hypermedia).

Chachoua, S., 2019. Contribution à l’évaluation de

l’apprenant et l’adaptation pédagogique dans les

plateformes d’apprentissage : une approche fondée sur

les traces. Environnements Informatiques pour

l’Apprentissage Humain. Université de La Rochelle,

2019. Français. NNT: 2019LAROS003. tel-02318386.

El-Kechaï, N., Muratet, M., Yessad, A. and Labat, J.-M.,

2015. Le suivi de l’apprenant : une approche fondée

sur le cadre méthodologique Evidence Centered

Design Application aux Serious Games. EIAH 2015-

Atelier Evaluation des Apprentissages et

Environnements Informatique, Jun 2015, Agadir,

Maroc. EIAH 2015.

Ennouamani, S. and Mahani, Z., 2017. An overview of

adaptive e-learning systems, Proceedings of the 8th

IEEE International Conference on Intelligent

Computing and Information Systems, 342-347, 2017.

Ezzat Labib, A., Canós, J. H. and Carmen Penadés, M.,

2017. On the way to learning style models integration:

a Learner's Characteristics Ontology, Computers in

Human Behavior 73 (2017), 433-445.

Farance, F., 2000.Draft standard for learning technology.

Public and private information (PAPI) for learners

(PAPI Learner). [Online]. Available: http://www.

iso.org/iso/iso_technical_committee%3Fcommid%3D

45392.

Felder, R. M. and Silverman, L. K., 1988. Learning and

teaching styles in engineering education. Engineering

Education, 78(7), 674–681, (1988).

Gaffney, C., Conlan, O. and Wade, V., 2014. The AMAS

Authoring Tool 2.0: A UX Evaluation, In Proceedings

of the 25th ACM conference on hypertext and social

media, pp. 224-230.

Ghallabi, S., Essalmi, F.,Jemni, M. andKinshuk, 2015.

Optimal composition of e-leaming personalization

parameters. In 5th International Conference on

Information andCommunication Technology and

Accessibility (ICTA).

CSEDU 2021 - 13th International Conference on Computer Supported Education

238

Heng, L. E., Sangodiah, A., Muniandy, M. and Yuen, P.

K., 2018.Integration of learner’s model into learning

management system environment, Journal of

Fundamental and Applied Sciences, 2018, 10(6S),

1771-1778.

Hlioui, F., Aloui, N. and Gargouri, F., 2016.A Survey on

Learner Models in Adaptive E-learning Systems.

IEEE/ACS 13th International Conference of Computer

Systems and Applications (AICCSA).

IMS Learner Information Package Specification version

1.0, 2001. [Online]. Available: http://www.ims

global.org/profiles/index.html.

IMS Reusable Definition of Competency or Educational

Objective version 1.0, 2002. [Online]. Available:

http://www.imsglobal.org/competencies/rdceov1p0/im

srdceo_infov1p0.html.

Kaya, G., Altun, A., 2011. A Learner Model for

Learning Object Based Personalized Learning

Environments. Communications in Computer and

Information Science, October 2011.

Lazarinis, F., Green, S. and Koutromanos, G., 2009. Web-

based learner profile management using IMS LIP.3rd

International Technology, Education and Development

Conference, pp. 1515-1521.

Li, H., Cui, W., Xu, Z., Zhu, Z. and Feng, M. Yixue, 2018.

Adaptive Learning System and Its Promise on

Improving Student Learning. In Proceedings of the

10th International Conference on Computer Supported

Education (CSEDU 2018), pages 45-52.

Madhour, H., Fernandes, E. and Wentland Forte, M.,

2006. Semantic Learning Model and Extended Student

Model: Towards An Aham-Based Adaptive System.

WBE 2006, Puerto Vallarta, Mexico.

Mediani, C., Abel, M.-H. and Djoudi, M., 015. Vers un

système de recommandation pour l’apprenant à partir

d’un modèle sémantique de connaissances dans un

environnement de collaboration. 7ème Conférence sur

les Environnements Informatiques pour

l’Apprentissage Humain (EIAH 2015), Jun 2015,

Agadir, Maroc. pp.180-191.

Radwan, N., 2014. An Adaptive Learning Management

System Based on Learner’s Learning Style,

International Arab Journal of e-Technology, Vol. 3,

No. 4, June 2014.

Ramírez-Noriega, A., Juárez-Ramírez, R., Martínez-

Ramírez, Y., 2017.Evaluation module based on

Bayesian networks to Intelligent Tutoring Systems, In

International Journal of Information Management 37

(2017) 1488–1498.

Peña, C., Marzo, J., and de la Rosa, J. L., 2002. Intelligent

Agents in a Teaching and Learning Environment on

the Web. In Proceedings of the international

conference on advanced learning technologies, 2002,

21-27.

Pesty, S, and Webber, C., 2004. The BagheraMultiagent

Learning Environment: An EducationalCommunity of

Artificial and Human Agents, In The European Journal

for Informatics Professional, Vol. V, No. 4, August

2004.

Tadlaoui, M. A., Aammou, S., M. Khaldi, M., and

Carvalho, R. N., 2016. Learner Modeling in Adaptive

Educational Systems: A Comparative Study.

International Journal of Modern Education and

Computer Science, vol. 8, n° 13, p. 1.

Tack, A., François, T., Ligozat, A.-L. and Fairon, C.,

2016. Modèles adaptatifs pour prédire

automatiquement la compétence lexicale d’un

apprenant de français langue étrangère. JEP-

TALNRECITAL 2016, Jan 2016, Paris, France. Actes

de la conférence conjointe JEP-TALN-RECITAL

2016.

Talon, B., Kerkeni, I., Tliche, S. and Belaïd Ajroud, H.,

2013. Tracking a collaborative Web 2.0 learning

environment, D.Hernandez-Leo et al. (Eds): EC-TEL

2013, LNCS 8095, pp 577-580, 2013. Springer-Verlag

Berlin Heidelberg 2013.

Tmimi, M., Benslimane, M., Berrada, M. and Ouazzani,

K., 2017. A proposed conception of the learner model

for adaptive hypermedia, International Journal of

Applied Engineering Research, ISSN 0973-4562,

Volume 12, Number 24 (2017) pp. 16008-16016.

Tnazefti-Kerkeni, I., Belaïd, H. and Talon, B., 2020.

An Adaptive Learning System based on Tracking, In

Proceedings of the 12th International Conference on

Computer Supported Education - Volume 2: CSEDU,

455-460, 2020.

Vandewaetere, M., Desmet, P. andClarebout, G.,

2011.The contribution of learner characteristics in the

development of computer-based adaptive learning

environments.Comput. Hum. Behav.27, 118–130

(2011).

Viccari, R. M., Ovalle, D. A. and Jiménez, J. A., 2007.

ALLEGRO: Teaching/Learning Multi-Agent

Environment usingInstructional Planning and Cases-

Based Reasoning (CBR), CLEI ELECTRONIC

JOURNAL, V CLEI ELECTRONIC JOURNAL,

Volume 10, Number 1, Paper 4, June 2007.

Wang, S., Feng, M., Bienkowski, M., Christensen, C. and

Cui, W., 2019. Learning from an Adaptive Learning

System: Student Profiling among Middle School

Students. In Proceedings of the 11th International

Conference on Computer Supported Education

(CSEDU 2019), pages 78-84.

Wartiningsih and Surjono, H. D., 2019. Adaptive E-

Learning Model in Learning Personality Characters,

Advances in Social Science, Education and

Humanities Research, volume 440, International

Conference on Online and Blended Learning 2019

(ICOBL 2019).

Yue Gong, 2014.Student Modeling in Intelligent Tutoring

Systems.Doctoral dissertation, Worcester Polytechnic

Institute, November 2014.

ADOPT: A Trace based Adaptive System

239