An Adaptive Learning System based on Tracking

Insaf Tnazefti-Kerkeni

, Henda Belaïd

and Bénédicte Talon

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

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

Keywords: Adaptive Learning, LMS, Personalization, Traces, Indicators, Agents, SMA.

Abstract: Success in training is an opportunity that must be offered to each student. However, many universities are

experiencing high rates of failure and dropout, especially during the first year of higher studies. We believe

that creating a process based on personalization of teaching can contribute to the decrease of failure rate

during undergraduate studies. To achieve this goal, we are specifically interested in online learning

supported by a Learning Management System (LMS). We have integrated, in a previous works, new tools

using traces of learners’ activities during collaborative works on an LMS. We therefore propose a system

based on intelligent agents. We are designing smart dashboards, automating detection of specific learners'

difficulties in order to offer alternatives or solutions to their problems.

1 INTRODUCTION

According to the study published by INSEE

(Dardier et al., 2013), learners' failure and

professional trajectories consisting of unemployment

and limited-term employment are often linked to

three main factors: history, educational level and

socio-demographic background, orientation during

the educational path.

The first factor – history - is linked to the nature

of the learning path (difficulties encountered,

changes in orientation, absenteeism, etc.). The

second factor – educational level and socio-

demographic - is dependant of the environment in

which the learner operates outside of study hours:

parental education, family income, etc. Finally, the

third factor – orientation during the educational path

- concerns specialties choosen during the school

career. This study also notices that "the Baccalauréat

holders registered in technological degrees drop out

less often than those registered in generalist

formation" (27% against 29%). According to the

study, individualized follow-up offered in

technological formations benefits to students.

Several reforms and actions have been

undertaken by the French state and other members

INSEE: Institut National de la Statistique et des Etudes

Economiques

of the OECD (eg: Multi-year plan against poverty

and for social inclusion launched). A slight drop in

the rates mentioned above is observed (ie: -3 points

on the rate of leavers without a diploma in 2015, -

0.8 points of the rate of young people without a job,

or training at the OECD level in 2015) . This

progress is due to a more regular and personalized

monitoring and to the integration of new

technologies in training and education services

(interactive platform, MooC for employment for

example).

The objective of our work is to design and

develop an effective model of adaptive learning. Our

model is designed to detect the learner profile and to

measure, in real time, the evolution of the learner's

skills. The aim is to permanently adapt the flow and

form of resources and to offer methods fitting the

needs and profile.

Two levels of individualization will therefore be

dealt with by the designed model:

1 - Recommendation on resources and

disciplinary contents.

2 - Recommendation on the type of support

(peer-learning, tutoring, etc.) adequate for the

detected profile.

The tool should offer monitoring functionalities

for teachers and training managers. These

functionalities are necessary to follow evolutions of

learners' skills and eventually prevent the risk of

dropping out or not completing the training. Alerts

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

An Adaptive Learning System based on Tracking.

DOI: 10.5220/0009571604550460

In Proceedings of the 12th International Conference on Computer Supported Education (CSEDU 2020) - Volume 2, pages 455-460

ISBN: 978-989-758-417-6

455

will notify actors (tutor, teacher, training manager,

etc.). Detailed situation reports (blocking points,

type of assistance needed, risk of dropping out,

recommendation for a solution, etc.) will allow

regular and individualized monitoring.

2 RELATED WORK

Researchers have been largely interested in

adaptation of educational systems. However, most

systems are used for specific context such as the

ELM-ART system (Brusilovsky et al., 1996). ELM-

ART is an adaptive hypermedia system designed to

learn the LIPS language.

More recently, we notice that ontologies are

largely used, as a basis for modeling in adaptive

systems. For example, in (Henze et al., 2004),

authors use semantic web to create learning

scenarios and to structure the courses. In

(Muruganandam et al., 2017), ontology is used to

model the learner profile.

A system called Manhali is presented in (El

Haddioui, 2015). It allows an adaptation of

educational strategies according to the behavior and

the learning style of the student. Based on the

learner's profile, adaptation is made on three levels:

graphic aspect of the platform according to the

configuration of the learner's machine, adaptation of

scientific content taking into account the skills of the

learner and adaptation of teaching strategies

according to behavior and style of the learner.

(Chachoua et al., 2016) uses traces left by

learners when they are active on an elearning

system. Duration of an activity and the number of

attempts to solve a problem are the traces that

interest the authors. These traces are used for

building an evaluation model. This model is

integrated in an adaptation model based on

ontological rules and an adaptation algorithm. The

result is adapted resources and learning strategy.

(Nafea et al., 2017) proposes an adaptive engine

that can be integrated into any LMS. It is based on

rules written in a rule-based reasoning algorithm.

3 ILLUSTRATION CASE

To illustrate our approach, a case study is presented

hereafter. It is adopted from a real situation we

encountered: G. is a first-year student at the

information Technology department of the IUT. He

starts his formation with 80 other students (each one

being "unique"). The training manager sends an

email to the educative team to inform about the

situation of this student. Here is a summary of this

email:

G. can only read with the help of his audio

device (and very hardly without). He hardly knows

how to write and has difficulties memorizing

information. He regularly forgets the meaning of

simple words and is not able to organize his ideas

and to express them. He has trouble doing

calculations and he can’t focus for long time. He

can’t stop moving or walking for a long time also.

Finally, G. may, under significant stress, be unable

to speak and to hear what is said to him.

So far, G. has benefited from a support service.

He is worried because he finds it hard to imagine

studying without this system that had allowed him to

have a certain balance in high school. However, G.

is courageous and of good will.

Although it is very complicated to make the

necessary adaptations for this student, the training

manager recommends that each teacher consider

this question and thinks about what to do for this

student in his teaching

G. will obtain (among other things) the right to

use his computer during exams as well as the right

to have the exams adapted (with reading and / or

reformulation).

This case perfectly illustrates the specific

conditions in which adaptation is needed to offer

ways to reduce the risk of failure of a student.

4 PROPOSED SYSTEM

We focus the Moodle LMS. We are working on

adaption of scenarios and contents in order to detect

specific learning difficulties of the audiences and to

best meet their specific needs (offering the adequate

learning scenario, the adequate resources in the right

form and a correct learning pace).

4.1 System Functionalities

A smart tracking system should then offer:

 Identification of learners who have a risk of

learning difficulties;

 Identification of the nature of problem (poor

assimilation of knowledge, unsuitable

resources - in content and / or format, etc.);

 Adaptation of the learning scenario, the

resources offered, the learning pace, the

support, etc. to each of these learners having

difficulties.

CSEDU 2020 - 12th International Conference on Computer Supported Education

456

So, the proposed system is based on two steps:

 Tracking Learners: creation and real-time

updating of a learner model to identify

students or learners who have problems or

risks. Detection of the nature of these

problems;

 Adaptation: presentation of alternatives or

specific solutions to these problems.

In a previous work (Talon et al., 2013), we have

developed a multi-agent system for tracking

students’ activities and calculating indicators. This

system collects various traces on the ILIAS learning

platform in the context of collaborative projects.

These traces are aggregated in the form of indicators

(of different categories). A dashboard allows the

teacher to assess activities of the students. It allows,

for example, detection of inactive students called

“sleeping students”. Dashboards form a space

where teachers can appreciate activity and

participation of each student but also offer to

students a way to be aware of their real activity. All

the developed indicators are purely informative.

Now, we go further by adding:

 “Traces on demand”. The teacher will choose

data he wants to collect and combine in order

to personalize a learning path.

 “Indicators formulation on demand”. The

teacher will create his/her own and specific

indicators. He/She will select elements in the

traces database and will determine the rules

that, according to him/her, should be applied

to formulate them.

4.2 A Multi-agents Platform

This tracking system uses agents to collect traces, to

update the learner model and to develop indicators

enabling the personalization of pedagogical

scenarios.

Agents are responsible for collecting traces

during the training of a student: his success, the

number of time he takes the tests to succeed, his

presence time on the LMS, the time he needed to

resolve a problem, etc. They also generate indicators

according to the teacher’s will, present them in the

teacher's dashboard and they are always looking for

the presence of conditions requiring the adaptation

of the scenario.

This system implies that the teacher:

 Defines and implements a "standard" scenario

in which he offers different resources

according to each cognitive ability (level 1 of

adaptability). Example: a common process to

develop the meta-competence VARIABLE

(knowing and understanding the concept of

variable, declaring a variable, different type of

data, etc.).

 Defines the indicators (on demand) that will

allow differentiating between students. For

example abstraction_level, is an indicator that

increases when a student fails in exercises

related to the notion of understanding what a

variable is. These indicators allow the teacher

to define rules. Thus, if the student has an

abstraction_level indicator less than 8, then he

should watch some specific videos before

passing the same scenario again.

If the adaptation of scenarios proves to be

unsuccessful, the system notifies the teacher who

can decide to modify the again or to deactivate them.

4.3 Learner Model

To face the needed adaptation of systems, data are

collected about learners and their activities in a

specific model called a learner model (Tack et al.,

2016).

Indeed, a learner model “consists of meta-

knowledge which includes the instructional

decisions about a learner” (Kaya et al., 2011). As

M. A. Tadlaoui and al. said in (Tadlaoui et al.,

2016), “the main objective of learner model is to

modify the interaction between the system and the

learner in a dynamic way to address the needs of

each learner on an individual basis”. So, the learner

model is necessary to adapt the learning process to

individual learning needs. As it is said in (Gong,

2014), “the student model is the core component in

an ITS”. In (Tmimi et al., 2017), it is judged to be

useful in the phase of learning and adaptation in an

adaptive hypermedia.

Y. Gong (Gong, 2014) developed the idea that a

learner model should integrate two elements: the

learner behavior when using the system and personal

properties such as the knowledge of the learner,

performance, etc.

To build a learner model, different methods

exist: cognitive science method, machine learning

method or both at the same time.

The data contained in a learner model are of

different types:

 Learner Data, which are personal information:

identity, gender, age, etc. gathered during the

registration process. It is a static view of the

learner model.

An Adaptive Learning System based on Tracking

457

Figure 1: System architecture.

 Pedagogical Data: Competency, knowledge,

program and subjects sequence.

 Personality Data: learning style, learner

features and preferences in learning. They can

be collected through tests conducted in LMS.

 LMS Experience: used to identify whether the

learner is familiar with a certain LMS.

 Cognitive Data: defines the preferences of the

learner. They can be obtained through tests

conducted in LMS.

We integrate all these information in the learner

model.

4.4 System Architecture

The personalization of learning is now initiated on

the multi-agent platform presented in the paragraph

4.2. In such a system, acting in a virtual environment

requires that the agents can make a representation of

the learning situation.

Agents must be able to have a representation of

the environment, the learner's task, the educational

actions to be carried out.

A first set of questions concerns the user’s

environment. What are the objects that constitute the

universe of the learner and where are they located?

What are their properties? What are the possibilities

of action on these objects? What behaviors can they

have? What interactions exist between these objects?

Concerning the representation of the learner's

task, it is necessary to know which actions the

learner is supposed to perform (and if there are

chained constraints) and which actions have been

performed (with success or not). An agent must be

able to explain what to do, possibly to do for or with

the learner. In a context of collective work, where

responsibilities are shared and defined by rules, it is

moreover necessary to know if it is the right person

who carries out an action.

The last type of questions concerns educational

interventions. An educational agent must be able to

give information on the learning situation

(accessible objects, task progress), but it must also to

modify the environment for an educational purpose:

in order to adapt to the learning, it may be necessary

to simplify the problem, by masking some elements

or by inhibiting some interactions.

Figure 1 presents the system architecture. The

different agents present in the system are the

following:

 a-observer: It tracks every action done by the

student when using the LMS. The raw traces

are cleaned and treated to get modeled traces;

 a-learner: It presents some indicators on the

dashboard of the student. These indicators

help the student to see his difficulties and to

get feedbacks regarding his learning;

 a-expert: It is a smart agent. It analyzes the

modeled traces to update the learner model. It

suggests adaptations to the teacher. Four

adaptation levels can be proposed:

o Navigational level proposes an order of

educational sequences

CSEDU 2020 - 12th International Conference on Computer Supported Education

458

o Content level proposes adequate

resources according to the student’s

knowledge and in relation with the

corresponding sequences

o Presentation level determines the better

form and nature of the resources

o Learning process level which defines

specific learning methods to adopt

during sequences

 a-teacher: It presents to the teacher tables and

graphs offering a monitoring space and

calculate indicators that will be displayed on.

5 CONCLUSIONS AND FUTURE

WORK

In this paper, we proposed an agent-based

Personalized Learning Architecture. The system is

characterized by the following properties:

 A learner model to store and permanently

update learner’s profile.

 Learning strategies according to the learner's

profile.

 Scenarios chosen by the course manager based

on prerequisites and learner's profile.

Ontologies play an increasing role in the new

generation of information or knowledge-based

systems. It is also a keystone of multi-agent systems

using high-level communication (Freitas et al.,

2017).

Our work is in progress. It consists firstly in

finalizing the ontology of the learner model.

Secondly agent integration and personalization of

scenarios will be dealt in the Moodle environment.

Our challenge is to identify, from traces and

questionnaires deployed throughout learning

processes on Moodle, the common trajectories

leading in achievement of objectives, and in

academic and professional success.

REFERENCES

Brusilovsky, P., Schwarz, E., Weber, G., 1996. ELM-

ART: An intelligent tutoring system on World Wide

Web. In Intelligent Tutoring Systems, Lectures Notes

in Computer Science, Vol 1086.

Brusilovsky, P., 2001. Adaptive hypermedia, User

Modeling and User Adapted Interaction, 11(1/2),

pp87-110.

Chachoua, S., Tamani, N., Malki, J., Estraillier, P., 2016.

Towards a Trace-Based Adaptation Model in e-

Learning Systems. In The 24th International

Conference on Computers in Education, Nov 2016,

Mumbai, India. hal-01899912.

Dardier, A., Laïb N., Robert-Bobée I. 2013. Les

décrocheurs du système éducatif : de qui parle-t-on ?

in France portrait social, édition 2013 INSEE Paris

El Haddioui, I., 2015. Manhali : un système de gestion

d’apprentissage adaptatif pour la modélisation du

comportement et la détection du style d’apprentissage

de l’apprenant. In 7ème Conférence sur les

Environnements Informatiques pour l’Apprentissage

Humain (EIAH 2015), Jun 2015, Agadir, Maroc.

pp.462-0. hal-01405992.

Freitas, A, Panisson, A., R., Hilgert, L., Meneguzzi, F.,

2017. Applying ontologies to the development and

execution of Multi-Agent Systems. In Web

Intelligence, vol. 15, no. 4, pp. 291-302,2017.

Gong,Y., 2014. Student Modeling in Intelligent Tutoring

Systems, Doctoral dissertation, Worcester Polytechnic

Institute.

Henze, N., Dolog, P. Nejdl, W., et al., 2004. Reasoning

and ontologies for personalized e-learning in the

semantic web. In Journal of Educational Technology

& Society, 7 (4), 82–97.

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

Object Based Personalized Learning Environments,

Communications. In Computer and Information

Science, October 2011.

Muruganandam, S., Srininvasan, N., 2017. Personalised e-

learning system using learner profile ontology and

sequential pattern mining-based recommendation, In

Int. J. Business Intelligence and Data Mining, Vol. 12,

No. 1, pp.78–93.

Nafea, S. M., Siewe, F., He, Y., 2017. An adaptative

learning ontological framework based on learning

styles and teaching strategies. In Proceedings of 85th

ISERD International Conference, Cairo, Egypt, 11th-

12th September 2017.

Parmentier P. 2018. La réussite des étudiants, objectif de

la transformation pédagogique. In Journées nationales

de l'Innovation pédagogique dans l'Enseignement

supérieur. 20 Novembre 2018

Tack, A,François, T., Ligozat, A.-L. Fairon, C., 2016.

Modèles adaptatifs pour prédire automatiquement la

compétence lexicale d’un apprenant de français langue

étrangère. In JEP-TALNRECITAL 2016

, Jan 2016,

Paris, France. Actes de la conférence conjointe JEP-

TALN-RECITAL 2016. <hal-01631772>.

Tadlaoui, M. A., Aammou, S., Khaldi,M., Carvalho, R. N.,

2016.Learner Modeling in Adaptive Educational

Systems: A Comparative Study.InInternational

Journal of Modern Education and Computer Science,

vol. 8, n° 13, p. 1.

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

2013.Tracking a collaborative Web 2.0 learning

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

TEL 2013, LNCS 8095, pp 577-583. Springer-Verlag

Berlin Heidelberg.

Tmimi, M.,Benslimane, M., Berrada, M., Ouazzani, K.,

2017.A Proposed Conception of the Learner Model for

An Adaptive Learning System based on Tracking

459

Adaptive Hypermedia. In International Journal of

Applied Engineering Research ISSN 0973-

4562Volume 12, Number 24, pp. 16008-16016.

Wenger, E., 1987. Artificial intelligence and tutoring

systems. Computational and cognitive approaches to

the communication of knowledge. San Francisco, Ed

CA: Morgan-Kaufmann.



CSEDU 2020 - 12th International Conference on Computer Supported Education

460