| March 2007 Index | Home Page |
Editor’s Note: Intelligent agents can facilitate learning in a number of ways. Here it is applied in a collaborative environment to achieve a significant increase in cognitive learning. In some ways the Système d’Apprentissage Collaboratif basé sur le modèle d’Agent (SACA) approach reflects the concept of learning objects. This research demonstrates an approach to developing learning materials that can be used at different levels and applied to a wide range of subject matter.
Supporting Learner’s Activities
in a Collaborative Learning System
“Système d’Apprentissage Collaboratif basé sur le modèle d’Agent”
Yacine Lafifi, Tahar Bensebaa
Algeria
Abstract
The aim of this paper is to present the main interactions carried out between the artificial agents of a collaborative learning system called SACA in order to support the learner’s activities. SACA is based on agent model in which learners collaborate to learn the concepts’ knowledge of the subject to be taught and to resolve the assessment exercises. These interactions ensure the various tasks which the system provides to its learners: learning, assessment and collaboration between them. Each activity is dedicated to an artificial agent composing SACA. This paper shows results of the experiment done at Guelma University.
Keywords: intelligent agent, collaborative learning, CSCL, interaction, collaboration, learning, assessment, pedagogical objective, tutor, pedagogical agent.
Introduction
Collaborative learning is a learning strategy where several learners interact with each other in order to achieve their common goals. Its impact on learner’s level is ensured; it is obvious that it is necessary to be interested in learning group environments instead of individual learning environments (Okamoto, & Inaba, 1997). The systems that support such strategy are called Computer-Supported Collaborative Learning (CSCL) system. CSCL is a new emergent paradigm which spreads out classical Intelligent Tutoring System (ITS) by introducing the concept of collaboration. It is then right not to make a difference between CSCL and ITS, but rather see group learning environments as a natural extension to individual learning in ITS (Okamato, & Inaba, 1997).
Many CSCL have already seen the day (Santos, Borges, & Systems., 1999; Lonchamp, 2006). Unfortunately, most of theses systems do not take into consideration the real need of the learner such as his preferences and mainly his level of knowledge during the collaboration. We suggest then, while collaborating, to take into consideration the aptitudes and the needs of the learner in order to offer him the possibilities of an effective collaboration, i.e. a collaboration which aims to improve the learner’s level and his capacities. Our work has achieved an implementation of a system called SACA (French acronym of “Système d’Apprentissage Collaboratif basé sur le modèle d’Agent”). The latter is an agent-based collaborative learning system that facilitates the learning process and the collaboration between different learners. Besides, it enjoys all the opportunities of intelligent tutoring systems. Artificial agents in SACA interact between them in order to ensure the following activities: learning, assessment and collaboration between learners.
The field of intelligent agents has been rapid growth over the last decade and such agents now constitute powerful tools that are utilized in most applications (Kim, Kim, & Rim, 2003). The main features of agents (as well as the modularity, the adaptivity and the autonomy) can make them good tools for designing collaborative learning systems.
Architecture of SACA
Structure of the subject to be taught in SACA
The subject to be taught is made up of a set of concepts regrouped into Pedagogical Objectives (PO). These correspond to a mental structure, an abstraction and are sometimes represented by conceptual networks. The teacher (instructor) can create pedagogical objectives, assign them a difficulty level and establish "prerequisite" relations between them (Lafifi, 2000). Each PO (X) can have a set of prerequisite pedagogical objectives that must be acquired by learner in order to learn the concepts of X (Bensebaa, & Lafifi, 2000).
In SACA, pedagogical objectives are represented by artificial agents called Domain Agents. These agents hold a set of information such as the knowledge represented by the pedagogical objective and the domain agents associated to their prerequisite pedagogical objectives.
Agent model in SACA
An agent is a computational entity that (i) executes in behalf of other entities (users, programs, etc.) in an autonomous way; (ii) makes actions in a pro-active and/or a reactive way; and (iii) presents some capabilities to learn, cooperate and move (Olguin, Delgado, & Ricarte, 2000).
Recently, various CSCL systems based on agent model have been developed. In these systems, “agents” with their own goals and functions are embedded, and perform their own tasks through communication and collaboration among them to achieve a goal as the system requires (Okamoto, & Inaba, 1997). We distinguish different kinds of projects working in multi-agent based learning environments. Some of them work on generic platform of agents but usually the focus is given to a specific agent type. Interesting results have been achieved by pedagogical agents regarding student motivation and companion agents acting sometimes as mediator of the learning process. Finally, tutor agents are usually related to student modeling and didactic decision taking (Webber, Bergia, Pesty, & Balacheff, 2001).
Among CSCL systems based on agent model, we can mention GRACILE (Ayala, & Yano, 1996), SHIECC (Labidi, Lima, & Sousa, 2000), SPLACH (George, 2000), Alice/WhiteRabbit (Blanchard, & Frasson, 2002), SIGFAD(Mbala Hikolo, 2003), I-Help(Vassileva, McCalla, & Greer, 2002.), etc.
In SACA, an agent is constituted of a set of modules and knowledge bases (see figure 1). Therefore, an agent possesses:
communication module: it allows the agent to communicate with other artificial agents in the system,
control module: based on a description of the agent‘s behaviours toward the messages that can be received from the other agents. It manages a knowledge base called behaviours knowledge base,
reasoning module: it uses the agent’s knowledge and a set of reasoning rules allowing it to accomplish its role in the system. It manages a knowledge base called reasoning knowledge base,
an optional module called interpretation module: associated to agents having an interaction with human actors (learner, teacher and tutor) because its main function is the interpretation of the human agents’ actions.
Multi-agents architecture:
Figure 1. Architecture of an artificial agent.
SACA is constituted of a set of artificial agents. Some of them are associated with system’s human actors. Each learner has the following agents:
An Assistant agent of Learner (AL): It proposes to the learner an interface which makes the learning task easier for him/her. It contains a learner’s student-model, his/her learning history and other information in its reasoning knowledge base.
Pedagogical Agent (PA): Its role is to present the pedagogical objectives to the learner according to his/her final profile and his/her current knowledge state. They are expressed by pedagogical objectives.
Collaboration Agent (CA): This agent takes into account the collaboration process between learners as well as the associated problems (interrupted collaboration, double collaboration, etc.). (Lafifi, & Bensebaa., 2006b).
Assessment Agent (AA): Its role is to measure the learner's knowledge level by proposing to him/her a set of exercises from various models and difficulties.
The teacher must initialize the assessment parameters and organize the subject to be taught as well as its structure in pedagogical objectives (set of concepts). To carry out these tasks, he/she has two agents:
An Assistant agent of the Teacher (AT): It proposes to the teacher an interface in order to assist him/her in the creation of the concepts and the exercises of the subject to be taught. Each type of exercises can test different kinds of knowledge. Among these kinds we can quote: «definitions», «correspondence between elements», «dependence degree», «methods and rules», etc. (Benadi, 2004).
A Mediator agent of the Teacher (MT): It facilitates the communication between the teacher and the learners or between teachers themselves (Lafifi, & Bensebaa, 2004).
In SACA, learners are organized in groups where they are helped and followed-up by human tutors. Each human tutor has an artificial agent called Agent of the Tutor which assists him/her in the realization of assigned tasks: giving councils to learners and following-up their learning processes (Lafifi, & Bensebaa, 2006a).
Interactions between the different agents of SACA
Figure 2 shows the various interactions between some agents of SACA. These agents support the following activities: learning (AL), assessment (AA) and collaboration (CA).
Request for self assessment (concerning a PO).
Presentation of exercises.
Answers of exercises.
Request for the learner’s assessment (concerning a PO).
Assessment’s result of a pedagogical objective.
Cognitive profile of learner.
Result of self assessment + cognitive profile.
Collaboration (using the various mechanisms of collaboration).
Selected pedagogical objective (to learn) + request for councils.
Pedagogical objective to learn + councils.
Following-up the learners + councils.
Figure2. Interactions between artificial agents in SACA.
Presentation of some interfaces
The human actors of our system are the learners, the teachers and the tutors. To each one of them is associated an interface. It is via the teacher interface that pedagogical objectives as well as assessment exercises are built. Each exercise must carefully be thought and proposed as a stage in a pedagogical progression (Govaere, 2000).
Interface of Learner
He/she can learn the concepts of the subject to be taught, self-assess or collaborate with the other learners. During his navigation, the learner moves from a concept to an other of the same agent of the domain or between the elements of knowledge of the same concept by disconnecting some links (the knowledge is presented in a form of hypermedia) (figure 3). A set of tools is offered to learners to save their states of advancing and their ways already covered.
Figure 3. Structure of the subject to be taught in SACA.
In SACA, the learners collaborate by using synchronous or asynchronous tools. These tools are:
Forums: we have implemented three types of forums: public forum (concerned all learners), group forum (for each group), and subject forum (for each subject to be taught),
Electronic mailing,
Semi-structured interface and Chat (figure 4).
In each tool, the learner can save the steps of the collaboration process (list of the sent and the received messages)(Lafifi, & Bensebaa, 2006c).
Figure 4. Collaboration tool “chat”.
Teacher interface
The teacher is the first responsible on the creation of the pedagogical objectives and the exercises. For this, he/she uses some tools that make theses tasks asier for him/her (see figure 5).
Figure 5. Creation of the concepts (of a PO) of the subject to be taught.
Tutor interface
The tutor follows-up the learners by giving to them advices and councils. He/She uses the forum by group to communicate with the learners of each group. Furthermore, he/she can see the cognitive and the social profile of each learner who belongs to his/her groups (figure 6 describes the main interface of the tutor).
Figure 6. The tutor interface.
Experiment
Methodology
An experiment has been done at Guelma University (Algeria). The learners were from computer science department, level: 2nd year students (forty four students). These learners learned a subject called “Algorithms II” which is composed of a set of six pedagogical objectives that contain about ninety concepts. The teacher created, for each one of the concepts, a set of exercises from different types and models. These exercises belong to the following models: "Question with Multiple Choices", "True or False", "Correspondence list", "Fill in the blank", "Simulating an algorithm", "Algorithm mistakes detection", "Classification" and "Open answer with only one word".
The participants were divided into two groups (at random). The first group (control group) follows a system prototype without collaboration between learners while the second (experimental group) follows a system prototype with all the functionalities. All the learners are organized in groups followed up by six human tutors from two departments: computer science department and mathematics department. Learners access to the system using the intranet of the university (rooms of practical works at the department of computer science, internet room at the university, etc.). At the end of the experiment (after three months), a questionnaire is submitted to the learners of both groups.
Our hypothesis is that “collaboration increases the cognitive level of learners”. In other words, the collaboration between learners, for resolving exercises or learning the subject’s concepts, increases their cognitive levels.
Results and discussion
To verify our hypothesis, we have compared the means of control group and experimental group. To know if the difference is significant between the two means we have used paired samples t-test (student t-test) (because the size of the sample is less than 30). After using R software (http://www.r-project.org/) which is a free software environment for statistical computing and graphics, we have obtained the following results with 95% as significant level (α=0.05):
Table 1
t-test statistics
N | Mean of control group | Mean of experimental group | t score | Degree of freedom | P-value |
22 | 8.773 | 10.32 | -2,9897 | 21 | 0.006985 |
From the table of t-test, t0.975=± 2.04, so tscore<t0.975 (-2.98 <-2.04) the difference was very significant, so the hypothesis is proved and we can affirm that “collaboration” can increase the cognitive level of learners in collaborative learning system.
As a general observation concerning the experimental group, we can say that:
86.36 % of learners collaborated between them.
Through some exercises, the learners acquire new knowledge (like “simulating an algorithm” exercises).
Faced problems
According to the students, the most frequent problems they meet are:
Some pages of the system seem to be full of information (especially those concerning collaboration). There are a lot of information on the same page (13 students agree on this problem).
No possibility of saving parts of the subject or solutions of exercises.
The tool help is qualified as insufficient by the majority of students.
Conclusion and Future Work
The choice of the intelligent agents for the modeling of our collaborative learning system (SACA) is promoter. The interactions carried out between artificial agents of SACA make it possible to provide an environment adaptable to the cognitive level of learners (good, weak, etc.), to ensure a fine assessment of each learner (by extracting the acquired knowledge and the not acquired knowledge) and finally to facilitate collaboration between learners (by providing some mechanisms and tools of collaboration (chat, forum…)).
The agents of SACA collaborate to support the various activities of learners:
structuring the knowledge to be presented to the learners,
following-up the learners,
assessing the acquisition of the learners’ knowledge,
taking into account the collaboration between the learners,
The short period of experimentation of the system has shown the interest of the application of such strategy of collaborative learning on the cognitive and social level of learners. The final marks obtained by learners and the collaboration rate between them (86%) validate the choice of such strategy in the educational field.
A very significant results resulting from this experimentation show the effectiveness of the interventions of SACA’s agents for supporting the learners’ activities: collaboration agent (looking for a good collaborator), assisting Agent of learner (interventions and councils), etc. Interactions carried out between these agents increased the quality of the services provided to the learners, which make the processes of learning, assessment and collaboration more beneficial.
For future work, we plan to develop the collaborative resolution of exercises (problems) by attributing different roles to learners (moderator, supervisor, etc.) and to develop the negotiation rules used in the case of conflict between learners.
References
Ayala, G., Yano, Y., 1996. Intelligent Agents to Support the Effective Collaboration in a CSCL Environment. Proceedings of ED-MEDIA & ED-TELECOM 96. Boston, Mass, USA.
Bensebaa, T., Lafifi, Y., 2000. Architecture d’un hypermédia éducatif et coopératif. Proceedings of 5ème Colloque Africain sur la Recherche en Informatique CARI, Madagascar.
Benadi, S.A., 2004. Structuration des données et des services pour le télé-enseignement. PhD thesis, Institut National des Sciences Appliquées de Lyon, France.
Blanchard, E., Frasson, C., 2002. Une architecture multi-agents pour les sessions d’apprentissage collaboratif. In Frasson C., Pécuchet H-P. (Dir), technologies de l’information et de la communication dans les enseignements d’ingénieurs et dans l’industrie. Villeurbanne INSA de Lyon, France. pp 283-287.
George, S., 2001. Apprentissage collectif à distance. SPLACH un environnement informatique support d’une pédagogie de projet. PhD thesis, University of Maine, France.
Govaere, V., 2000. Evaluation et guidage d’un utilisateur dans un environnement d’apprentissage, application au domaine de la rééducation de la parole. PhD thesis, university of Henri-Poincare, Nancy 1, France.
Kim, D.S., Kim, C.S., Rim, K.W., 2003. Modelling and designing of intelligent agent system. International Journal of Control, Automation, and Systems, Vol 1, N° 2.
Labidi, S., Lima, C. M., Sousa, C. M., 2000. Modelling Agents and their Interaction within SHIECC: A Computer Supported Collaborative Learning framework. The International Journal of Continuous Engineering and Life-Long Learning. Special Issues on Intelligent Agents for Education and Training System.
Lafifi, Y., 2000. Architecture d’un hypermédia éducatif et coopératif. Master thesis, Annaba University, Algeria.
Lafifi, Y., Bensebaa, T., 2004. SACA : un système d’apprentissage coopératif. Proceedings of SETIT’04, Tunis.
Lafifi, Y., Bensebaa, T., 2006. Supporting collaboration in agent-based collaborative learning system. Proceedings of EEE ICTTA’06, Damascus, Syria.
Lafifi, Y., Bensebaa, T., 2006. Evaluation paramétrable dans un système d’apprentissage collaboratif. CEMAFORAD 2006, Sousse, Tunisia, November 12-15.
Lafifi, Y., Bensebaa, T., 2006. Outils pour favoriser une collaboration effective dans SACA. Proceedings of MajecStic 2006, Lorient, France.
Lonchamp J. 2006. Supporting synchronous collaborative learning: A generic, multi-dimensional model. International Journal of Computer supported collaborative learning, Vol 1, Issue 2.
Mbala Hikolo, A., 2003. Analyse, conception, spécification et développement d’un système multi agents pour le soutien des activités en formation à distance. PhD thesis, University of Franche-Comté, France.
Okamoto, T., Inaba, A., 1997. The Intelligent Discussion Coordinating System for Effective Collaborative Learning. Workshop Notes IV, Artificial Intelligence in Education.
Olguin, C.J.M., Delgado, A.N., Ricarte, I.L.M., 2000. An agent infrastructure to set collaborative environments. Educational Technology& Society 3(3).
Santos N., Borges M. R. S., Systems C. 1999. Computer supported cooperative learning Environments : A framework for analysis. Proceedings of EDMEDIA / ED-TELECOM 99. World Conference on Educational Multimedia & Hypermedia and Telecommunications. Seattle, Washington, USA, June 19-24.
Vassileva, J., McCalla, G., Greer, J., 2002. Multi-agent multi-user modelling in I-Help. User modelling and user adapted interaction (2002) E.Andre and A.Paiva (eds.), Special issue on user modelling and intelligent agents.
Webber, C., Bergia, L., Pesty, S., Balacheff, N., 2001. Baghera project: a multi-agent architecture for human learning. Multi-Agent Based Learning Environments workshop, AIEd 2001, San Antonio.
About the Authors
Yacine Lafifi is a researcher at Guelma University and he prepares for his PhD Thesis at Annaba University in Algeria. He is interested in collaborative learning, assessment of learners, e-learning and CSCL.
Address: LAIG laboratory, Guelma University, BP 153
Guelma Maouna, Guelma 24000, Algeria,
laf_yac@yahoo.fr
Tél: 00213 37 21 67 63
Fax: 00213 37 206 8 72
Tahar Bensebaa is an associate professor at Annaba University. He obtained the PhD thesis in 1991 from INSA lyon, France. He is the head of a research group at LRI laboratory in Ananba University. He is interested in Hypermedia, Collaborative learning, Pedagogical Simulation, Computer assisted instruction, CSCL and E-learning.
Address: Computer science department, Annaba University, BP 12
Sidi-Ammar 23200, Annaba, Algeria