AN INTELLIGENT ASSISTANT TO SUPPORT STUDENTS
AND TO PREVENT THEM FROM DROPOUT
Tri Duc Tran
1,2
, Bernadette Bouchon-Meunier
1
Christophe Marsala
1
and Georges-Marie Putois
2
1
LIP6 DAPA, Université Pierre et Marie Curie, 104 Avenue du Président Kennedy, Paris, 75016, France
2
ILOBJECTS, 104 Avenue du Président Kennedy, Paris, 75016, France
Keywords: Non pedagogical agent, Intelligent Personal Assistant, Educational Agent, Tutor, Coach, Student support,
School help, Dropout prevention.
Abstract: The research and development of an educational agent are commonly focused on the pedagogical aspect; the
main objective is to automate the teaching activity, to replace the teacher in virtual learning environment.
Our goal is different: it is to create a non pedagogical intelligent assistant that follows students during their
learning to prevent dropout. This assistant can be compared to a coach, a motivator, or a mentor that
motivates, encourages, and helps students to overcome their difficulties.
1 INTRODUCTION
Nowadays the use of Information Technology (IT) is
essential in education. The development of IT in
learning is one of the most important preoccupations
for the future.
Our aim is to use Artificial Intelligent
mechanisms as machine learning, fuzzy logic, and
intelligent agents to build a personalized and
autonomous virtual assistant that helps students
during their learning experience. Currently the
intelligent agent technology is broadly used in e-
learning with a pedagogical purpose; it helps
students to learn and intends to replace teachers. The
aim of our approach is not to build another virtual
teacher; the services of our assistant will be focused
on the non pedagogical aspects of learning such as
the management of motivation, stress, school
orientation and task organisation. The actions of our
assistant will be oriented at complementing the
teacher’s work.
In this article we present our concept of an
assistant supporting student learning in three
sections. The first section will explain the concept of
educational agents, in proposing a taxonomy of
educational agents. In the second section we will
study the different types for student support: tutor,
coach, and mentor. And the last section will focus on
the architecture of our intelligent assistant.
2 A TAXONOMY OF
EDUCATIONAL AGENTS
2.1 Characteristics of Educational
Agent
The analysis of the taxonomy of educational agent
indicates the positioning and the functionalities of
our assistant. Educational agents can be considered
as “Software agents” (Franklin&Graesser, 1996) and
more precisely “User agents”. This kind of agent is
based on the concept of delegation and indirect
management tasks (Sanchez, 1997); agents offer to
end-users a new approach to interact with computer
systems.
The Sanchez’s taxonomy (Sanchez, 1997) can be
extended in adding a new type of agent. The
Intelligent Personal Assistant (IPA) is a sub-type of
User Agents. We consider that an IPA differs from
Information Agents, Task Agents or Synthetic
Agents because it is more personalized and its
relationships with its user are closer and durable. It
can be considered as virtual companion.
The role of an IPA is to reduce the complexity
and the rigidity of human-machine interactions, and
to anticipate the needs of the user with some
personalization capacities (Briot&Demazeau, 2001).
The Foundation for Intelligent Physical Agents
167
Tran T., Bouchon-Meunier B., Marsala C. and Putois G. (2009).
AN INTELLIGENT ASSISTANT TO SUPPORT STUDENTS AND TO PREVENT THEM FROM DROPOUT.
In Proceedings of the First International Conference on Computer Supported Education, pages 166-171
DOI: 10.5220/0001848701660171
Copyright
c
 SciTePress

(FIPA) states that a personal assistant is like a
secretary, it accomplishes routine support tasks to
allow the user to concentrate on the real job, it is
unobtrusive but ready when needed, rich in
knowledge about user and work (FIPA, 2000). The
personal assistant will work in collaboration with the
user in the same environment; the most important
part of a personal assistant is the management of the
user profile (Maes, 1994).
IPA can have the appearance of a simple
software interface with button, textfield, list,
radiobutton… Or it can be personalized with a
human representation, a multimodal user interface as
an Embodied Conversation Agent (ECA) that can
carry a conversation with the user through the
common communication modalities like speech,
gestures, body stance, and facial movements.
The virtual companion, IPA represented through
an ECA can be refined in various kinds depending
on its purposes. It can be used for educational
assistance, professional helps for complex task or
entertainment. Educational agent is a specific kind
of Embodied Conversation Agent (ECA) with a
representation in 2D or 3D and a natural language
communication capacity.
In our approach, an educational agent is a
software agent, user agent, Embodied Conversation
Agent and Personal Intelligent Assistant (see Fig. 1).
Figure 1: An extended software agents’ taxonomy.
From the extended taxonomy (see Fig. 1), we can
determine the common and essential capabilities of
educational agents as it inherits the properties and
capabilities of Intelligent Personal Assistant and
Embodied Conversation Agents.
Capacities of Intelligent Personal Assistant
(Maes, 2003), (Sanchez, 1997):
 To assist a user to perform task so it can hide
the complexity
 To have a certain degree of reasoning and
autonomy; user can delegate some tasks to the
agent.
 To manage the user model and the domain
model
 To learn the user’s interests, goals and
preferences from the interaction of the user
with the IT system or with the intelligent
assistant. This capacity is the most important
and it is essential for personalization.
Capacities of Embodied Conversation Agent
(Cassel et al., 2000):
 To recognize and respond to verbal and
nonverbal input
 To generate verbal and nonverbal output
 To deal with conversational functions such as
turn taking, feedback and repair mechanisms
 To give signals that indicate the state of the
conversation and contribute new propositions
to the discourse
The design of an educational agent has to include
at least all of the previous capacities. In the next part
we will explore the capacities of different kinds of
educational agents.
2.2 Different Types of Educational
Agents
In the education field, the use of intelligent agents
can increase the attention of the student and make
the learning more attractive. An Intelligent Personal
Assistant should be a key success factor for online
learning tools. Experiments conducted in
California’s university (Baylor, 2003) showed that
the use of pedagogical agents motivate students and
facilitate the learning. Usually, the educational
agents have a pedagogical goal, their presence is
supposed to replace the teacher in an e-learning
environment.
(Chou et al., 2003) highlight two types of
educational agents and place them into two
categories:
 pedagogical agents involved in learning
activities, they simulate a teacher. ITS
(Intelligent Tutoring System) is an example of
this type of agent.
 personal assistants providing help and
information that pertains to learning activities,
like collecting content to perform an activity
or reminding tasks.
CSEDU 2009 - International Conference on Computer Supported Education
168

The exploration of the taxonomy of educational
agents permits to define, identify and explain the
general role of our assistant. We can see that there is
a lack in the development of non pedagogical
agents; a lot of the educational agents on the e-
education market are principally designed to
improve the transfer of knowledge without an
intervention of human teacher.
The taxonomy of educational agents determined
by Chou et al. (2003) only focuses on pedagogical
agents, and how to improve the content learning
transfer. But in the learning activities or processes
there are other aspects. For example, motivation is
essential in cognitive learning processes (Barnier,
2003) (Pintrich, 2000). Students need to be
motivated and encouraged when they are in a failure
situation. The non-pedagogical aspects of learning
can include:
 the management of stress,
 the management of emotion : fear, anxiety,
reduce the attention and cognitive abilities,
 a time management, task planning, scheduler,
 a vocational orientation,
 the identification of strengths and weaknesses,
 a development of self-confidence,
 a development of responsibility,
 a development of autonomous.
We can add four types (see Fig. 2) of student’s
agents to the description of pedagogical agents from
(Baylor, 2003) and (Ali, 2002):
 the digital tutor
 the digital secretary
 the motivator agent
 the mentor agent
Figure 2: Different types of student’s assistant in
educational agents’ (Chou and al., 2003) taxonomy
refined.
Mentor agents and motivator agents can also be
considered as pedagogical agents but here we will
only retain the non-pedagogical aspects of these
agents.
Our assistant is a personal assistant and is in the
sub-type student’s assistant. In the next chapter, we
will explore the concept of a tutor, a coach and a
mentor and precise the role of our non pedagogical
assistant.
3 DIFFERENTS TYPES OF
STUDENT’S SUPPORT
With the massification of education, school dropout
is one of the most important challenges for the
education system. The best solution to help students
to prevent dropout is the personalization of the
learning to fit the student’s abilities. A personalized
support relation will allow improving their results.
3.1 Dropout: The Main Issue in 21st
Century’s Education System
In OCDE (Organisation for Economic Co-Operation
and Development) countries, between 5% and 40%
of students drop out of school and 30% of adults
have only primary or lower secondary school (Field
et al., 2007).
The consequences of failures at school are
multiple:
 It generates stress, anxiety, and decrease self-
confidence for students. These can have
serious consequences for the teenagers like
depression, behavioural disorders (runaway,
flight, fight …).
 It increases the costs of education system; the
financial costs of educational failure are high.
The causes of dropout can be differentiated into
three categories of factors (Roiné, 2007):
 Individual factors: tardiness, suspension,
absenteeism, lack of motivation, low social
network, poor or trouble relationship with
adults, disciplinary infractions, low self-
esteem, substance abuse
 Familial factors: family organization, parental
responsibility, socioeconomics status, poor
education of parents
 School factors: relationship between teacher
and student, equity in education, negative
school climate.
Our assistant will focus on individual factor to
help students. In the next section, we present
AN INTELLIGENT ASSISTANT TO SUPPORT STUDENTS AND TO PREVENT THEM FROM DROPOUT
169

different types of support or help to prevent school
failure. The different kinds of help can be classified
by the relation with the learning content (Shea,
2004), the master just transmits a learning content
while the mentor transmits the life experience and
assists in the decision making.
3.2 Tutor
The concept of tutor is mainly encountered in e-
learning environments; his mission is to maintain the
motivation of learners. The tutor is the link between
the e-learning system and the learner. His three main
competencies are (Denis, 2003):
 pedagogical and relational: evaluates student
works, analyse the progression, help and guide
students
 technical: gives advise for technical problems
 disciplinary: gives academic subjects
instructions and resources
3.3 Student Coaching
The role of a coach is to help student to identify and
exploit his potentials. The study of French’s school
coaching shows that the helps are focused on
vocational orientation on management of
motivation, stress and self confident.
3.4 Mentor
It’s the most personalized and closed relationship in
learning. This type of relationship is commonly used
in enterprise for the career path and the turnover
management. A mentor is model, a motivator and
advisor for students (Houde, 2004).
3.5 The Supporting Services
The analysis of non-pedagogical agents and
supporting methods has permitted to identify four
types of services for our non pedagogical assistant:
student coaching, vocational guidance, virtual
secretary and technical help (see Table 1).
The student coaching, vocational guidance
modules are designed in using psychology concepts,
affective computing and cognitive learning.
The technical part can be considered as an
Intelligent Help System (Winkels, 1992) which
assists the user with a current problem and to teach
the user about the information system.
Table 1: Our four types of services.
Services Functions
Student coaching Diagnoses difficulties
Motivates when student has bad
results
Gives a concrete meaning to the
learning
Increases the self confidence
Reduces stress and anxiety
Vocational
guidance
Gives a feedback on the potential
of the student
Identifies the potential vocation
appropriate to the student’s
strengths and weaknesses
Virtual Secretary Reduces cognitive overload
Reminds tasks
Organizes and plans work
Manages files and contents
Technical help Helps to use new tools
Guides in the Virtual Learning
Environment
4 THE ARCHITECTURE OF OUR
ASSISTANT
In the previous sections we have identified the issues
and the roles of an intelligent assistant to support
students and to help them to overcome difficulties
during learning.
4.1 Architecture of Our Assistant
The architecture of our assistant has four main
modules. It can be considered as an instantiation of
the ITS’s architecture (Wenger, 1997) without the
dimension of pedagogical expertise (see Fig. 3).
1) User Profile. This module manages all the
information about the user, his environment and the
Information Technology platform (virtual learning
environment, e-learning tool). Technologies of
semantic information representation can be used as
ontology and the norm OWL to design user’s data
representation.
2) Embodied Conversation Agent. This module is
the human-computer interface; the interaction with
the user is in natural language.
3) Reasoning. This is the core of our assistant;
fuzzy-logic based inference mechanisms are used,
and machine learning tools are added to detect when
a student encounters difficulties. The early
identification of the failure is the key success factor
CSEDU 2009 - International Conference on Computer Supported Education
170

of our system. The reasoning module contains the
student’s diagnosis and the non pedagogical helps
modelling.
4) Interoperability Technology. This is an interface
that manages the exchange of information with a
Virtual Learning Environment or e-learning portal
and the integration of our assistant. This is the IT
interoperability layer of our system.
Figure 3: Architecture of our intelligent supporting
assistant.
4.1.1 User Profile Module
The goal of the user’s profile is to store student’s
characteristics. The management of user description
is essential to an adaptive and personalized system.
Its objectives are:
1) To construct a model of student characteristics
2) To capture and maintain the coherence and
consistency of the student profile
3) To give the pertinent indications to the HMI
(Human Machine Interaction) and Reasoning
modules.
The tracking of interaction with the Embodied
Conversation Agent, and the use of the keyboard and
mouse will allow the construction of the student
profile. The user model is composed of four sub
modules. The Basis, Environment, and Domain
models are based on the works of (Brusilovsky,
2001), and (Kobsa, 2001).
 The Basis model maintains personal
information as interests, preferences; cognitive
profile, learning style and schooling’s data:
identification, personal data, interest,
preferences learning results.
 The Affective model manages the student’s
emotion through the tracking of keyboard and
mouse interactions. It will detect the stress, the
sadness (depression) and the anger. This
model is based on the theory of affective
modelling by (Picard, 1997).
 The Environment model keeps information
about the student’s work context: type of
device, place, and time.
 The Information System (IS) model handles
the information about the technical
environment in which the assistant is
integrated, for example a Virtual Learning
Environment. And the information about the
student’s learning, for example the school’s
organisation.
4.1.2 Embodied Conversational Agent
Module
The interaction between student and assistant is
carried out by an animated agent. First the
communication will be in text mode like chatting on
instant messenger and later we can add speech
capacity.
The assistant can have the appearance of a
talking head or a full-body character (see Fig. 4) and
is considered as a multimodal system.
Figure 4: Different types of animated agent: full-body
character and talking head agents.
Its objectives are:
1) To manage the dialog model to communicate with
the student, it may include the facial or body
corporal expression,
2) To interact with the user profile to learn new
knowledge about student.
4.1.3 Reasoning Module
This module has two objectives, the first one is the
detection of the weaknesses and the difficulties of
the student and the second is to offer an appropriate
help. On ITS the student’s diagnosis aims to detect a
specific lack of knowledge, in our case the global
difficulties on school subjects are diagnosed.
The detection of the student’s difficulties is
based on the values of grades from learning
evaluations, tardiness, suspension, absenteeism
AN INTELLIGENT ASSISTANT TO SUPPORT STUDENTS AND TO PREVENT THEM FROM DROPOUT
171

(excused or unexcused, frequency), disciplinary
infraction, teacher’s remark. The evaluation of
theses values individually and jointly will determine
the degree of failure and activate the helping pattern.
5 CONCLUSIONS
In this paper we have studied the functionalities of a
non pedagogical intelligent assistant to support
students during their school learning though the
taxonomy of educational agents and different types
of the student support. This assistant can be
considered as a coach or a companion that gives
psychological helps and advices when student
encounters difficulties. Our aim is to reduce the
dropout through the use of Artificial Intelligent
methods.
The limitations of our system reside in:
 The monitoring of the current student’s state:
school performance, physiological and
psychological state. The performance level of
our system depends on the quality of the
analysis of these states.
 The construction of expert knowledge to
support student.
 The relation between the assistant and the
student: relevant and acceptance of advices
from the intelligent assistant.
To solve these difficulties we need to combine
Artificial Intelligence techniques with psychology,
cognitive and learning theories.
Until now we have conceived the architecture
and lead some experimental development of each
module separately:
 The conversation module is build with a
pattern-design model based on xml,
 The diagnosis of dropout is based on fuzzy rule,
 The user’s profile is designed with web
semantic standard; OWL is used to maintain
user’s characteristics.
Our next step is to make these modules work
together to build a prototype of our non pedagogical
agent, test and validate it in real-world applications.
REFERENCES
Ali, J. 2002. Conceptualizing Intelligent Agents for
teachning and learning. Educause Quartely, Number
3.
Barnier, Gérard. Théories de l’apprentissage et pratiques
d’enseignement. IUFM d’Aix-Marseille, 2003.
Baylor, A. 2003. The impact of threee pedagogical Agent
Roles. In AAMAS’03, July 14–18, Melbourne,
Australia.
Briot, JP., Demazeau, Y. 2001. Principes et architecture
des systèmes multi-agents. Collection IC2, Hermès.
Brusilovsky, P. 2001. Adaptive Hypermedia. User
Modeling and User Adapted Interaction, Vol. 11, pp.
87-110.
Cassell J., Bickmore T., Campbell L., Vilhjalmsson H.,
Yan H. 2000. Conversation as a system framework:
Designing embodied conversational agents. Embodied
Conversational Agents. MIT Press.
Chou, C., Chan, T., Lin, C. 2003. Redefining the learning
companion: the past, present, and future of
educational agents. Comput. Educ. 40, 3 p.255-269.
Denis, B. 2003. Quels rôles et quelle formation pour les
tuteurs intervenant dans les dispositifs de formation à
distance. Distances et savoirs, Hermes, CNED,
Lavoisier 2003 p.19-46.
Field, S., Kuczera, M., Pont, B. 2007. No More Failures:
Ten Steps to Equity in Education . OCDE.
Franklin, S., Graesser, A. 1996. Is it an Agent, or just a
Program?: A Taxonomy for Autonomous Agents. In
Proceedings of the Third International Workshop on
Agent Theories, Architectures, and Languages.
Springer-Verlag.
Houde R. (2004). Le mentorat, une culture à consolider.
Le Devoir (Montréal).
Kobsa A. (2001): Generic User Modeling Systems. User
Model. User-Adapt. Interact. 11, p. 49-63.
FIPA. 2000. Personal Assistant Specification.
http://www.fipa.org
Maes, P. 1994. Agents that Reduce Work and Information
Overload. In ACM Communications, Vol. 37, n°7,
pp.30-40.
Picard, R., W.1997. Affective Computing. MIT Press.
Pintrich, P.R. 2000. The role of goal orientation in self-
regulated learning. In Boekaerts, M., Pintrich, P.R., &
Zeidner, M. (Eds.), Handbook of self-regulation (pp.
451-502), San Diego: Academic Press.
Roiné, C. 2007. La psychologisation de l’échec scolaire :
Une affaire d’état. In Congrès International AREF.
Sanchez, J. A. 1997. A taxonomy of agents. Tech. Rep.
ICT-97-I. Laboratory of Interactive and Cooperative
Technologies. Department of Computer Systems
Engineering. Universidad de las Americas-Puebla.
Wenger, E. 1987. Artificial Intelligence and Tutoring
Systems: Computational and cognitive approaches to
the communication of knowledge. Los Altos: Morgan
Kauffmann Publishers.
Winkels, R.G.F.: Explorations in Intelligent Tutoring and
Help. IOS Press, Amsterdam (1992).
CSEDU 2009 - International Conference on Computer Supported Education
172