A system based on workflows and Agents
Latifa Mahdaoui, Zaia Alimazighi
University of Sciences and Technologies ‘Houari BOUMEDIENE’ - Algeria
Faculty of Electronic and
Computer Science
Computer System Department; LSI
Laboratory – ISI Team
: El Alia BP n°32, Bab Ezzouar, Alger, Algeria.
Keywords: Process; activities; task; former; formation; role; workflows; multi-agents; e-learning
Abstract: In the E-learning platform, we can consider three principal actors (Teacher, Learner and Administrator)
whose inter
act or cooperate between them among processes, then in context of enterprise the e-learning
process can be seen as a cooperative information system where actors are managers and employees. Many
of these processes can be automated and then we consider this work as a workflow process. The learning
process is naturally flexible because of different levels of learners and the different ways to present a lesson
or training process. We use an oriented object Meta-Model based on UML to describe a process concerning
Tutor and Learner and we propose a Multi-Agent System (MAS) based on ITS architecture to support the
work of the actors roles “tutor” and “learner”.
The e-Learning made its appearance following the
technological developments of information and
communication meanings such as the Web and the
Internet. Issued from distant teaching, the first
platforms of e-Learning consisted in providing tools
and convivial interfaces to three principal actors of
the system : Teachers, Students and Administrator,
allowing essentially :
For teacher : creation of courses
, incorporation of
teaching multi-media resources (sound, image,
video) and more or less a follow-up of learners.
For learner: consulting courses on line or
ownloading contents, resolving exercises,
transmitting duties for correction.
For administrator: management and control of
and learners; management of pedagogical
resources and technical maintenance of the system.
Among platforms, we find GANESHA which is
very sim
ple or others more elaborated like Web-CT
and Web-Tutor (improved tutoring quality). But, in
majority of cases, the recommended solutions are
directed more towards management of the teaching
contents than the teaching itself. In other words, the
question ishow to make sure a good quality of
teaching and an effective follow-up of learning in
absence of direct interaction with teacher ?”. In
traditional classroom, teacher can play several roles.
Among these, the tutor role. In fact, our interest is to
perceive the e-learning with a different vision
regarding the teaching contents built by the teachers
or the specialists as necessary resources for training
So, the training process is seen as a transfer of
owledge according to a given pedagogical
structure with a planning of work. The learner will
have to carry out the planned work in order to ensure
a good knowledge acquisition.
Consequently, the training process appears as a
ccession of activities that the tutor will define and
the learner will perform throughout his studies.
Hence, the idea is to design e-learning process based
on cooperation between tutor and learner.
Each platform’s actor can have different ro
according to objectives of teaching. In one side, the
tutor will have also to perform some activities for
the follow-up of his learners group. Each actor will
need a whole resources in order to achieve suitably
the tasks which fall onto him. In the majority of
cases, these resources are of various document types
such as courses, exercises, multi-media contents, …
etc. Then the concept of route appears. Finally, the
execution of tasks will have to be accomplished
according to a certain number of pre-established
Mahdaoui L. and Alimazighi Z. (2005).
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents.
In Proceedings of the Seventh International Conference on Enterprise Information Systems, pages 213-225
Roles, rules and routes are the basic characteristics
of workflow processes (Khoshafian & Al, 1998).
Therefore, we propose to design e-learning process
based on workflow (for the processes corresponding
to this category well-sure) (Cesarini & Al., 2004)
(Lin & Al.1, 2002) (VanTroys & Al., 2002) (Lin &
Al.2, 2002). We are particularly interested to
learner’s following-up by the tutor. To do that, we
use a meta-model approach designed using UML
(Alimazighi & Al., 2002) (Saikali, 2000).
The advantage of using meta-model is to allow the
elaboration of working plans for learners with
different profiles. Each instance of model (workflow
case) will concern a learner and his tutor. Learners
can evolve according to their own speed and the
tutor can perform a personalized tutoring for each
one of them.
On the other side, we are not always teaching in the
same manner and all learners have not the same
skill’s level; so, the learning process is naturally
flexible. For this, using workflow meta-model may
allow to do changes at the instance workflow level
and the model workflow level. Thus, a tutor can
operate modifications on an instance of a particular
learner. In this case we talk about “instance
flexibility”. If changes concern all the model
instances, then we talk about “model adaptability”
(Saikali, 2000). The meta-model cover all the
aspects of the workflow process (organizational,
functional, behavioural and informational aspects).
The behavioural aspect of the system is expressed
using extended UML activity diagrams
Workflow processes are asynchronous; so, tutors
and learners are relatively free to perform their work
regarding their needs. This leads to the absence of
direct interaction between the tutor and the learner
and constitutes a lack for good teaching quality as
the learner don’t have any way to obtain immediate
assistance. Therefore, we propose a Multi-Agents
System (MAS) based on cognitive agents to support
tutors and learners (Garro & Al., 2003) (Pesty & Al.,
2001). This paper is structured as follows :
Section 2 : in this part, we locate the e-learning
process regarding to the four aspects of the
workflow meta-model : organizational, functional,
behavioural and informational aspects.
Section 3 : we design an example of e-learning
process and discuss about points which necessitate
assistance for a best tutoring quality.
Section 4 : this section describes MAS’s agents
among their roles and their functionalities. A general
overview of possible interactions between agents
themselves and human actors is also described.
Section 5 : summarizes the obtained results and
future works.
Within the framework of our research’s team, a
meta-model of workflow process was developed
(Alimazighi & Al., 2002) in accordance with the
standards of the WFMC (http://www.wfmc.org)
which is a consortium of workflow standards. The
meta-model covers four aspects :
Organizational aspect : describes the organizational
structures, the actors of the system and their roles.
Functional aspect : shows system’s functionalities
by splitting of a process into sub-processes,
activities and tasks.
Behavioural aspect : enhances the control flows,
conditions and events attached to the activities.
Informational aspect : presents the part of
information system necessary to the achievement of
Initially, we locate the e-learning process relatively
to these four aspects.
2.1 Organizational aspect
Many organizations of training and education use
today Internet for distant formation. In our point of
view, a platform of e-learning can be seen as an
additive support to university’s campus (Mahdaoui
& Al.1, 2004). This could contributes to reduce the
loads endured in term of capital costs. In fact, the
installation of an e-learning platform (equipments,
networks, software tools,…etc), could not be higher
than costs of real infrastructures (construction of
buildings, classrooms, human resources, …etc). We
consider then e-learning platforms like a virtual
institution of teaching whose actors and roles are :
Table 1: Actors and roles in e-learning platform
Actors Roles
Teacher Tutor
Author of contents
Student Learner
Administrator System Manager
Manager of Teachers and
Manager of pedagogical contents
Let’s note that Administrator roles and examiners
are not concerned by this study. The Administrator
is assumed to be the system. For considered
organisational structures, we propose to dispatch
learners which adhere in e- learning formation into
groups. Each group will have a tutor by module or
topic taught within the framework of the followed
formation. Parameters such as the number of learner
per group or a number of groups per tutor are fixed
by the administrator (the tutor is concerted). For
example, one tutor can propose an evaluation test to
detect learner’s profile in order to form groups.
The roles described for the teacher can be ensured
by the same or different persons. We consider that
the role concept is independent from the person who
plays it. The following figure shows the organisation
of actors and roles (fig. 1) :
Figure 1: Organization of actors in e-learning process
The following diagram presents the part of the
meta-model covering the organisational aspect (fig.
Figure 2: Organizational Meta-Model
2.2 Functional aspect
The functional aspect describes processes in term of
sub-processes, activities and tasks independently of
rules, events and constraints to which they are
subjected. Thus, by regarding to the planning of
work established by tutor as a process, we notice
that it can be broken up into activities and each one
decomposed into tasks to be achieved by the learner.
We consider the task as the smallest entity of work,
i-e, not decomposable. The functional part of the
meta-model is presented in the figure below (fig. 3):
Figure 3: Functional Meta-Model
2.3 Behavioral aspect
This aspect focuses on intrinsic flow of control to a
process and allows to show the state of activities and
tasks. We can also describe conditions and events
stream controlling the process execution. We
suppose that the necessary pedagogical resources
(tools and support documents) are disposable for the
work of learner at each moment.
The activity diagrams of extended-UML
(http://www.omg.org) allow to describe these
properties. The description of a process always
begins with the word “START” and finishes with
the word “END”. We use swimlane’s activity
diagrams where each swimlane represents a role
(tutor or learner) and contains the set of activities or
tasks to perform. A complete example will be shown
in section3. Figure 4 illustrates the behavioural
aspect :
Figure 4: Behavioral Meta-Model.
Transitions express the activities and tasks. An
activity or a task is characterized by one or several
possible states. An activity is described by a
succession of executable tasks. If all activity tasks
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents
are correctly performed then the activity is correctly
accomplished. The links between transitions
describes relations of precedence, parallelism,
choice, … etc. A transition may be triggered by one
or several events and submitted to simple or
complex conditions. This can be described by
following operators : OR-SP, OR-JO, AND-SP,
AND-JO, and XOR-SP as shown below (fig. 5, 6
and 7).
Figure 8: Iterative Structure
2.4 Informational aspect
Finally, the informational aspect presents the
information system part in order to extract necessary
data for operation of learning and then the execution
of e-learning workflow processes. Information can
be portions of databases or any other documents
(text, multi-media, formulary, …etc). Meta-model
part covering this aspect is represented in fig. 9 :
Figure 5: OR Operator
Figure 9: Informational Meta-Model
To extract pertinent information for this part, we
carry out a conceptual study based on UML and
RUP (Rational Unified Process) (Jacobson & Al.,
2000) (http://www.omg.org). We obtain a class
diagram covering all the information of e-learning
platform. From this diagram, we present only
information concerning this work, like showed in
figure 10 by the dotted framework:
Figure 6: AND Operator
Figure 10: Classes Diagram for the Informational aspect
Figure 7: XOR Operator
2.5 Integration of the four aspects
Let‘s note however that the XOR-JO operator is
similar to the OR-JO. Activity can executed
iteratively as showed in (fig. 8):
The diagram of figure 11 shows the integration of
the four aspects previously introduced. For
clearness, the information system part is presented
without details.
Figure 11: Workflow meta-model for e-learning
3.1 Presentation of the example
In this section, we will present an example of e-
learning process model by designing the
correspondent activity diagram according to the
behavioral meta-model. We consider that a learner’s
work plan is prepared by the tutor. To do that, he
retrieves structured contents (courses, exercises, …
etc) established by the author. Tutor can also insert
his own exercises. Furthermore, he can consult
domain’s specialists and pedagogues to improve
learner’s work plans.
Let’s consider a module of “Information System”
(a process) where a part of program is described in
(fig. 12):
Figure 12: Decomposition of a process into sub-processes activities
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents
The granularity choice in terms of processes, sub-
processes, activities and tasks is delegated to the
tutor. In this example, “Information System”
represents the whole process subdivided into courses
represented by sub-processes which are split into
multiple activities corresponding to lessons. Each
activity contains a succession of tasks to be
performed by a learner. Let’s note that nothing
prevents to see a process differently. For scheme’s
legibility, we use zoom on sub-processes and
activities as illustrated in figure 13:
Figure 13: Zooming of a sub-process and activities
In correspondence with learner’s work plan, the
tutor will define a set of tutoring activities. This
depend on his judgment of critical points (for
example, need of learner’s feed-back) in the work
plan. Figure 14 shows an example of tutoring plan
for sub-process1 (see fig. 13) :
Figure 14: Example of tutoring process
3.2 Activity diagram describing the
An activity diagram specifies the global process
behavior and the work part of both actors learner and
tutor. Events and control flows are detailed and
learner-tutor interaction is enhanced. Tasks due to
interactions are added like “Send a course” for tutor
and “Connect to learning session” for learner. See
figure 15 for the enrolling of sub-process1 (of fig.
Figure 15: Activity diagram for a sub-process1 learning session
For clearness, we have reduced our example to
a normal scenario, but in fact, the procedure can be
more complex. We have said in introduction that e-
learning is flexible, so when the tutor conceives a
work plan, he can prevents possible assistance
points besides pertinent points (necessary for
tutoring). Here, process modeling becomes more
complex and if he (the tutor) don’t want to do it, he
can ignore exceptional cases and treat them during
workflow case (flexibility). If the tutor detects that
the same exception arises at the same point in all
workflow cases then he can decide to modify the
process model (adaptability). That is possible by
using of meta-model (Saikali, 2000)
(http://www.wfmc.org). Figure 16 presents an
example of learner’s assistance for a task.
Figure 16: Learner’s assistance for the task “Study the
examples and case of study
: Tas
: Activity : Even
[…] : Condition
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents
We differentiate between learner’s connection to
platform and the execution of e-learning instance
process. Therefore, when a learner is connected to
the system, he must identify himself as adherent to
the platform (administrative side). Once the
connection is established, he selects the module he
wants to study and then accedes to activity list (sub-
process). Interactions between tutor and learners are
asynchronous, hence it’s possible that a learner
stands until his tutor sends the results of tutoring
before to continue his work (for instance, he wait
for an answer to his sent question). Finally, a
workflow case concerning one matter (module) can
be performed in several sessions of connection, as
it’s possible that different sub-processes instances
for the same module are executed in one connection,
depending on learner’s rhythm, skills and the
process model.
From there, the using of workflow system is
benefic in sense that it makes possible to plan both
works of tutor and learners. Asynchronous
communication allows each one to evolve
accordingly to his disposal. But the learning quality
may be altered because some types of problems that
the learner can meet need immediate response (this
is not possible when the tutor is not connected).
In the other side, a tutor have one or several
groups of learners. They can solicit him at the same
time for different problems from various degrees of
importance, so tutoring can become difficult.
Thus, to improve the quality of learning and
tutoring, both roles need assistance. We propose to
introduce artificial actors in the system. These actors
are organized as a Multi-Agents System (MAS)
(Garro & Al., 2003) (Ouahrani & Al., 2003) (Pesty
& Al., 2001).
4.1 Needs of assistance for tutors and
As previously described, tutors and learners need
assistance during the accomplishment of their
respective tasks. These can be summarized in
For learner, the primordial thing is to provide
an assistance or an artificial tutoring (as much
as possible) during his work’s session.
For tutor, the essential is to present and
organize all critical information concerning
his groups in such way he can takes a good
and efficient tutoring decisions.
To realize that, we propose the intervention of
artificial actors in the system. These actors are
cognitive agents having abilities of reasoning and
taking decisions (Garro & Al., 2003) (Ouahrani &
Al., 2003). They can communicate between them
and with human actors by messages. Agents can also
execute complex tasks alone or in
cooperation(Ouahrani & Al., 2003). Each agent play
a specific role in the system. The MAS is organized
accordingly to Intelligent Tutoring Systems (ITS)
(Mahdaoui, 2002). Advantages of using MAS in this
context are :
To make tutoring so easy by delegating all
automatisable tasks to agents. This will
allows tutor to focus on real problems which
can not be resolved without him.
To offer a certain necessary interactivity for
learner to evolve, and to allow more rapid
progress in training by submission to
elaborated learning strategies (Aimeur & Al.,
1999) (Mahdaoui, 2002).
The MAS can collect lot of knowledge and
this may considerably contributes to the
improvement of tutoring and learning
strategies. Tutoring strategy is the “tutoring
way” of tutor and the learning strategy is the
manner of learner’s learning.
Finally, let’s note that when adding a MAS, our
goal is not the elimination of human tutoring but its
improvement with best level of assistance.
Organizational meta-model of workflow is extended
to agents like showed in figure 17:
Figure 17: Organisational Meta-Model extended to agents
4.2 MAS’s Agents description
Agents are defined by the role they can play in the
system. The MAS is constituted by seven cognitive
agents : “Learner’s Assistant Agent” (LAA),
“Personal Tutor Agent” (PTA) and “Pedagogical
Agent” (PA) are dedicated to learner. “Tutor’s
Assistant Agent” (TAA) and “Group’s Assistant
Agent” (GTA) concern tutor. “Super Tutor Agent”
(STA) and “Session Manager Agent” (SMA) are for
both tutor and learner. We describe below, each
agent and its functionalities.
4.2.1 Learner’s Assistant Agent (LAA)
The role of LAA consists to assist each learning
session. It’s a learner’s representative towards the
other agents concerning him. One LAA agent is
associated to one learner.
Functionalities :
9 LAA is created when a learner is connected
for the first time to the e-learning platform.
At each learner’s connection, it’s activated by
the system and inactivated at each
9 LAA helps the learner in the choice of the
module to study. After, it informs PTA agent
that the learner is on-line.
9 When learner wants to obtain help from
others, LAA make in his disposal a special
buffer where he can write his request.
Moreover, the learner can precise the receiver
of message which can be PTA agent, the tutor
or other learners in his group. If the receiver
is human then LAA demand to SMA if he
(human) is connected.
9 If SMA’s reply is well acknowledged then
LAA negotiates with the corresponding agent
of the receiver (other LAA or TAA for the
tutor) to know if he wants to converse. After
answer reception, LAA inform the learner.
9 If the learner have not precise the receiver
(for example, he search for any connected
learner in his group), then LAA demands a
list of connected learners from SMA and
after, LAA negotiates to find favorable
persons to dialogue. It transmits results to the
9 Furthermore, LAA can accede to learner’s e-
mail once it is activated and classify arrived
and non treated messages. Tutor’s messages
are always placed at the top of the list.
Generally, LAA can use the message list to
council the learner for choice module.
4.2.2 Personal Tutor Agent (PTA)
PTA is an artificial personal tutor for the learner and
assists him for each task he performs. For each
learner, one PTA is associated. PTA knows the
nature and pedagogical objectives of activities and
tasks. We distinguish between two categories of
tasks : passive task like reading text or visualizing
multi-media and active task like doing exercises,
duties or practical works.
For passive tasks, PTA will inspect parameters
like estimated duration for example. This kind of
task remains very hard to inspect, so, PTA will just
try to attract learner’s notice that it must be
Generally, an active task always terminates with
learner’s feed-back. Activities like exercises, for
example, are particularly important for a good
teaching and presents pedagogical goals as :
Test of acquired knowledge
Increase learner’s capabilities to analyze and
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents
Evaluate learner’s skills and lacks
9 PTA receives a sub-process (work plan)
from GTA and loads learner’s work-list
with the activity to execute. Each task of an
activity corresponds to a work-item and
PTA will assist its achievement.
9 If the task is passive, the learner can decide
to jump it (for example, because he knows
the content) and so PTA will note this fact.
After, PTA can use it to direct his tutoring
9 If the task is active, PTA will stand for
learner’s feed-back. If necessary, PTA can
activate his subordinated agent PA for
applying an appropriate pedagogical
strategy to the learner. When terminated,
PTA will receive a report from PA.
9 PTA knowledge allows it to judge if a task
is correctly performed or no (if learner’s
results are satisfactory or not). Then PTA
can take one of the following decisions :
If results are good then PTA will pass to the
next step of work. It can be the next task or
activity to do. If the concerned sub-process
is correctly finished then PTA will request
from GTA the next work to be done.
Depending on GTA’s answer, PTA may
continue tutoring or arrest the work (answer
depends on what the tutor have arranged).
In arrest case, PTA sends a detailed report
to GTA about the learner.
If results are not good, PTA will try to
resolve the problem according to the
following steps :
¾ It proposes for learner to re-execute the task
in the step which causes the problem.
¾ If problem persists (not conclusive results),
PTA will use his own experience and
propose other works (complementary tasks)
to reinforce learner’s knowledge.
¾ If PTA have not required competences or if
results remain not good, GTA will be
solicited for help.
¾ GTA’s reply can be : a new tutoring
strategy or information that the problem
will be transmitted to the tutor. In this case,
PTA stops tutoring until reception of new
Moreover to that, PTA can cooperate to resolve
GTA’s requests for other PTA’s count. If PTA have
no response to send, it ignore the request.
4.2.3 Pedagogical Agent (PA)
PA’s role is to submit a learner to different learning
strategies in order to ensure that he had (the learner)
understand and for filling his gaps. PA is
subordinated to PTA which solicits it when
necessary. According to pedagogical objectives, PA
can become “companion”, “troublemaker” or
“learner’s pupil” (Aimeur & Al., 1999) (Mahdaoui,
9 PA can apply one of these strategies :
Learning with companion : PA
becomes a co-learner (a friend) with
slightly high level of knowledge than
learner. PA guides the learner during
his work by counsels and suggestions
(according to what it knows). The
learner is free to choice and to decide.
Learning by disturbing : PA becomes a
troublemaker. With this strategy, the
level of exercises is more elaborated
and aims to provoke conflict situations
(cognitive dissonance) between PA and
the knowledge and the convictions of
the learner (Aimeur & Al., 1999). So,
PA can detect learner’s gaps and his
performances in analyze and proof.
Learning by teaching : it’s the ultimate
step where PA becomes a learner’s
pupil, i-e, learner have a very high
level of knowledge and capacities.
Therefore, he must teach to PA.
9 Once terminated, PA transmits a report
containing evaluations, marks and
observations to PTA.
4.2.4 Tutor’s Assistant Agent (TAA)
Similarly to LAA, TAA represents the tutor over
against other concerned agents and assist him at
each tutoring session. For one tutor corresponds one
9 TAA is created when a tutor is connected
for the first time to e-learning platform.
TAA is activated/inactivated according to
tutor’s connection/disconnection.
9 Tutor can have several associated groups
for different modules. Then TAA assists
him for choice. After, TAA informs
concerned GTA that the tutor is on-line.
9 When tutor wishes to communicate with
agents or his learners, he describes his
request in a special area provided by TAA
with possibilities to choose receivers.
9 When tutor want to send his reply
concerning a problem received from GTA,
he also use TAA.
9 TAA treats the request like described for
9 TAA can accede to tutor’s e-mail once it is
activated and classify arrived and non
treated messages. Learner’s messages are
always placed at the top of the list and
generally, TAA can uses the message list to
council tutor for the group and the matter
4.2.5 Group Tutor Agent
GTA is associated for one group of PTA
representing a group of learners and one tutor (for
one matter). GTA is an artificial tutor for all group
members in sense that if any problem PTA(s) cannot
be resolved, GTA will try to do it. GTA cooperates
with PTA and STA and informs regularly the tutor
about group’s evolution.
Functionalities :
9 When GTA receives request from PTA, he
attempts to answer according to its own
experience (knowledge previously collected
from others PTA, STA or tutor).
9 If GTA don’t find the solution, it requests
SMA to obtain a list of active PTA(s)
(inactivated PTA(s) have been previously
sent all the information to GTA).
9 Based on the received list, GTA diffuses a
request for tutoring assistance. If no result
is obtained then GTA asks STA for help.
9 Once answers received, GTA analyses and
organizes results in a report and sends it to
concerned PTA. If no solution, GTA
informs PTA that a problem will be
submitted to the tutor.
9 GTA can receive request for assistance
from STA, thus if it cannot treats the
problem, GTA ignore the request.
4.2.6 Super Tutor Agent
STA capitalizes all the experiences concerning
GTA(s) that tutors the same module (for the same
profile of learners) and experiences concerning the
same group of learners for different modules. At
present, we consider that only one STA exists for all
the groups.
Functionalities :
9 When STA receives a help demand from
GTA, it try to treat it according to its own
knowledge (results from other GTA(s)
9 If no results, STA diffuses a request for
assistance towards actives GTA(s) (list of
actives GTA(s) is obtained from SMA).
The rest of the procedure is similar to
GTA/PTA scenario.
9 When help’s demand is received from the
tutor, depending on what is requested, STA
treats question under one of the two aspects
described in the beginning of this
4.2.7 Session Manager Agent
SMA has the responsibility to save all the
information concerning connection/disconnection of
learners and tutors (and activated/inactivated
agents). Actually, we consider one SMA for all the
system. SMA is automatically launched by the
system when start-up.
Functionalities :
9 Receives and treats demands issued from
all other agents of the MAS and tutors.
4.3 General Overview of the MAS
As previously described, figure 18 summarize the
global behaviour of the MAS :
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents
Communication Communication / Cooperation
Communication between agents of the same type Requests for services
Figure 18: Agents Interactions
The link between workflow system and MAS’s
agents is activities and tasks received in the learner’s
work-list and work-items. Agents may communicate
and cooperate between them. Communication with
human actors is also possible. To realize this system,
we are working to specify the MAS with AUML
diagrams suitably to the FIPA standards
(http://www.fipa.org). AUML allows specification
of agents (competences, knowledge, intentions,
plans …etc). Moreover, we aim to enhance the
conversation between agents and to define the
appropriate cooperation/ communication protocols
and to avoid conflict problems.
In this paper, we are interested to design an e-
learning process as a workflow process. Some
related works have considered workflow aspect in
the e-learning in different ways. (Cesarini & Al.,
2004) (Lin & Al.1, 2002) propose models for
teaching staff who interact between them.
(VanTroys & Al., 2002) (Lin & Al.2, 2002) propose
a workflow engine for e-learning based on
Workflow Management Facility (WMF)
(http://www.wfmc.org). (Pesty & Al., 2001) (Garro
& Al., 2003) talk about using MAS for teacher’s and
student’s partner system.
Our approach, proposes to combine the
using of workflow process with a MAS to improve
the quality of assistance for tutors and learners. We
propose to design both works of learner and tutor,
and the interaction between them by workflow
processes. We have established an UML meta-model
of Workflow process for e-learning. For the
behavioural aspect, extended UML activity diagrams
are used. These constitute a powerful formalism
allowing the expression of control flow using
specific operators.
We define a MAS containing seven cognitive and
cooperative agents. Like Intelligent Tutoring
Systems (ITS), some agents have the ability to
replace human tutor in determined tasks and help the
learner during his work. Knowledge capitalized in
the MAS can be used to improve the quality of
teaching in a platform.
As first result of implementation, we have
realized an author’s tool to prepare pedagogical
contents and exercises making it in disposal of tutor
for preparing work plans of learners. A tool for
definition of pedagogical workflow process is under
As perspectives, we are actually working on
specification of the behavioural aspect of workflow
with Workflow-Nets (issued from PETRI-NETS)
(Mahdaoui & Al.3, 2004). In fact, e-learning process
might be very complex and then the need of
properties verification is important before any
deployment on e-learning platform. Parallely to this
we have proposed in (Mahdaoui & Al.2, 2004) an
assistance system based on holonic agents an we
give a more detailed description of the MAS agents
using J. Ferber’s Grill and petri-nets in (Mahdaoui &
Al.4, 2005). Our future gool is to compare the two
approaches respectively to certain properties of
efficiency we will define.
In addition to what we said in section 4, we think
that the role of some agents can be extended in order
to cooperate with other actors like the administrator.
Finally, we want to consider other human roles like
examiner which represents a great importance in e-
learning platform as virtual structure of teaching.
This role can have important incidence in tutoring.
Aimeur, E., Fasson, C., Lalond, P. (1999) « Learning with
positive aspect of conflicts”. ACM Journal of Science,
Vol 27(3), 19 pages.
Alimazighi, Z., Hamdah, M. (2002) « Modélisation de
processus workflow avec le langage EXPRESS ».
CIP : Conférence Internationale sur la Productique,
Octobre 2002, Algeria.
Cesarini, M., Monga, M., Tedesco, R. (2004) “Carrying on
the e-learning process with a Workflow Management
Engine”, SAC’04, Nicosia – Cyprus. Copyright ACM
Garro, A., Palopoli, L. (2003) « MASel : A Multi-Agents
System for e-learning and Skill Management »
Calabria University, Italy (exp.telecomitalialab.com)
Vol 3 n° 3.
(2000) « Le processus unifié de développement
logiciel ». Editions Eyrolles, Juin 2000, ISBN
Khoshafian, S., Buckiewics, M. (1998) «Groupware &
Workflow», MASSON editions.
Joe Lin, Charley Ho, Wassim Sadiq, Maria
E.Orlowska (2002) « On workflow enabled e-learning
services ». University of Queensland, Australia.
Joe Lin, Charley Ho, Wassim Sadiq, Maria
E.Orlowska (2002) « Using Workflow Technology to
Manage Flexible e-Learning Services ». Educational
Technology & Society ISSN 1436-4522, pages 116-
123, 2002.
Mahdaoui, L. ; Alimazighi, Z. (2004) « Conception d’une
plate-forme de e-learning avec UML/XML et
implémentation d’un outil d’aide à la construction de
cours en ligne ». In proceedings of SETIT’2004.
Sousse – Tunisia 2004.
Mahdaoui, L (2002) “L’EAO, entre théories
psychologiques et nouvelles technologies ». 2
rencontre internationale sur l’université virtuelle,
Mahdaoui, L., Alimazighi Z. (2004) « Holonic Agent
System for workflow based e-learning». In
proceedings of the 3
European Conference on E-
Learning (ECEL 2004), November 25
- 26
, 2004.
Université Paris-Dauphine – France.
Mahdaoui, L., Boukhedouma, S. & Alimazighi, Z. (2004)
“designing e-learning process with workflow: an
approach based on UML and Petri nets”. In
proceedings of ACIT2K’2004 constantine Algeria.
Mahdaoui, L., Boukhedouma, S. & Alimazighi, Z. (2005)
“Vers un e-Learning à base de workflows et d’agents
cognitifs”. To be appear In proceedings of I S P
S’2005 Algiers Algeria.
Ole E.M.Borch, Jan Helbe, Morten Knudsen, Ole
Rokkjaer (2003) « UniFlex : A www environnement
for project-based collaborative learning ». Aalborg
University, Danemark. In proceedings of ITHET’03
July 7-9 2003, Morocco.
Ouahrani, L., Alimazighi, Z. (2003) « Agents pour la
coopération de systèmes d’information », Maghrebien
Conference, Tunisia.
Pesty, S., Webber, C., Balacheff, N. (2001) “Baghera : une
architecture multi-agents pour l’apprentissage humain.
Leibniz Laboratory, IMAG, France (www-
Sadiq, S., Sadiq, W., Orlowska, M. (2002) «Workflow
Driven e-learning beyond collaborative
environments». School of Computer Science and
Electrical Engeineering , Universiry of Queensland,
Saikali, K. (2000) “Flexibilité des workflows par
l’approche objet : 2FLOW, un framework pour
workflow flexibles”, Thèse de doctorat. Ecole
Centrale de Lyon – France.
Thomas VanTroys, Yvan Peter (2002) “A WMF-based
workflow for e-learning”. Université de Lille,
Laboratoire Trigone, Equipe NOCE, France.
A COOPERATIVE INFORMATION SYSTEM FOR E-LEARNING - A system based on workflows and Agents