Learning Management System Based on SCORM

0 downloads 0 Views 574KB Size Report
The current state of development of Learning Management Systems (LMS) is mainly based on ..... The learner attributes are used to compare the values of.
Learning Management System Based on SCORM, Agents and Mining Carlos Cobos1, Miguel Niño1, Martha Mendoza1, Ramon Fabregat2, and Luis Gomez3 1 Systems Department, University of Cauca, Colombia {ccobos, manzamb, mmendoza}@unicauca.edu.co 2 Institute of Informatics and Applications (IIiA), University of Girona, Spain [email protected] 3 Systems and Computing Engineering School, Industrial University of Santander, Colombia [email protected]

Abstract. Based on SCORM sequencing and navigation specifications, a learning management system has been developed. The system has intelligent tutoring system capabilities that allow contents, presentation and navigation to be adapted according to the learner’s requirements. In order to achieve that development, two concepts were put together: multi-agent systems and data mining techniques (especially the ID3 algorithm). All the implementation code was developed in VS.Net, which implied building a supporting framework for agents. The results of a pilot test were favorable. Keywords: Learning Management System, Intelligent Tutoring System, Adaptive Hypermedia Systems, SCORM, Agents, Data mining, ID3.

1 Introduction “Learning Management System (LMS) is a broad term that is used for a wide range of systems that organize and provide access to online learning services for students, teachers, and administrators. These services usually include access control, provision of learning content, communication tools, and administration of user groups. Another term that often is used as a synonym for LMS is learning platform. Two examples of well-known, commercial LMS systems are WebCT and Blackboard.” [1] The current state of development of Learning Management Systems (LMS) is mainly based on Sharable Content Object Reference Model (SCORM) specifications proposed by Advanced Distributed Learning (ADL) [2]. These specifications define a metadata schema for learning objects, a content package structure for courses and some sequencing and navigation rules (SN) [3]. SCORM SN defines both the learning activity concept and the sequencing rules for them. It also defines several smaller cluster units (the basic building blocks of learning activity [4]) and the content organization based on a tree structure called “Activity Tree” (AT) which can be useful as a Learning Instructional Template (LIT). Unfortunately, this AT is used by all learners regardless of particular skills or experience. Each student has his/her unique way of learning. A learning style is defined as characteristic strengths and preferences in the ways people take in and process information [5]. Many different learning style B. Benatallah et al. (Eds.): WISE 2007, LNCS 4831, pp. 298–309, 2007. © Springer-Verlag Berlin Heidelberg 2007

Learning Management System Based on SCORM, Agents and Mining

299

assessment models and instruments are available. In order to build an AT adapted to each learner, a support model for several learning models is essential. It is extremely ambitious for a teacher to build an AT by hand for each learner. Integrating Adaptive Hypermedia System (AHS) capabilities into an LMS facilitates the generation of personalized AT. The concept of adaptation has been widely investigated in the field of hypermedia systems [6]. “Adaptability” is the system’s capacity to adapt itself automatically to the user from its suppositions about the user’s necessities [7]. The developed LMS can adapt contents, presentations and navigation [6]. Data mining techniques are useful to analyze the learner profile (LP/user model) and to apply specific SN rules for each learner. In this paper, agent architecture and a data mining approach are proposed to construct an adapted AT with an associated SN for each learner. Agents work cooperatively to monitor and adapt every activity and content used by the learner based on his/her learning style, while a decision engine based on an ID3 implementation [8] creates a decision tree (DT) and an algorithm called “Delivering” generates an AT for each learner. In Section 2, related work is described and the added value of this work is explained. In Section 3, the main concepts and the relationships among them are presented in a conceptual model. In Section 4, the LMS architecture is presented to demonstrate the interaction of all agents and SCORM metadata with the decision engine. In Section 5, details of the agent’s implementation are shown. In Section 6, the data mining approach with ID3 and Delivering are proposed. In Section 7, the results of a pilot test are presented. Finally, Section 8 presents conclusions and future work.

2 Related Works Shang et al. [9] proposed an intelligent environment for active learning to support a student-centered, self-paced, and highly interactive learning approach. The learning environment can use the related learning profile (learning style and background knowledge) of students to select, organize, and present customized learning objects. Triantafillou et al. [10] proposed an adaptive learning system, called AES-CS (Adaptive Educational System based on Cognitive Styles), in which learners are divided into two groups, Field Independent and Field Dependent, according to their cognitive styles. The AES-CS system provides an appropriate strategy and learning objects for each group. Gilbert et al. [11] applied Case Based Reasoning (CBR) techniques to assign a new learner to the one group out of four with the most similar learning style. The system can offer adaptive learning material based upon the learning experience in the group selected. In all systems mentioned above, the approaches rely only on the adaptation of learning objects for each group. Weber and Specht [12] proposed an adaptive learning system called Episodic Learner Adaptive Remote Tutor (ELM-ART). It chooses the next best step according to the curricula presented to the learner. Links to pages are signaled following a “traffic light” metaphor that uses different colors to indicate recommended study

300

C. Cobos et al.

sections, learned sections or study sections not recommended because the learner is not ready to learn them yet. However, they do not monitor learner’s activities. Chen et al. [13] apply DT and data cube techniques to analyze the learning behaviors of students and discover the pedagogical rules of students’ learning performance from web logs including the amount of time spent reading article, posting article, asking question, login, etc. However, they do not apply educational theory to model the learning characteristics of learners, nor do they apply results to a new learner automatically. Peña et al. [14] propose an AHS called MAS-PLANG based on learning styles. Here, the adaptation techniques focus on customized learning object selection and navigation tools. MAS-PLANG models a learner with its own monitor agents. However, they use only one model (Felder and Silverman Learning Style Model [5]). Wang and Shao [15] propose a model called the Hierarchical Bisecting Medoids (HBM) Algorithm which integrates learner clustering, association mining techniques and historical navigation sessions by time. In the same group, the association mining technique was used to analyze navigation sessions and establish a recommendation model. In this approach, however, learning characteristics and student learning sequences were not considered, so personalized recommendation may not be appropriate. Finally, Su et al. [16] propose an adaptive learning system based on SCORM SN and a four phase Learning Portfolio Mining (LPM) approach, which uses sequential pattern mining, a clustering approach, and DT creation sequentially to extract learning features from learning portfolios and to create a DT to predict which group a new learner belongs to. However, they do not use a multi-agent system for system implementation or adapt learning objects. In this work, the automatic adaptation of learning objects, sequences and navigation rules classifies students according to their own learning styles while also Table 1. Differences between this work and related work Project vs. Characteristic

[9] [10] [11]

Adaptation of for each learning objects group Monitor learner’s MAS activities SCORM No supported active learning/ Based on education theory cognitive styles Apply results to a Yes new learner automatically Multiple learning No models Data mining No techniques DT: Decision Tree

[12]

[13]

[14]

“traffic light” metaphor

No

No

[16]

This work

object historical selection and navigation navigation

No

for each learning style

No

MAS

No

No

MAS

No

No

No

No

SN

metadata, SN

No

No

learning models

No

No

learning models

Yes

No

Yes

Yes

No

Yes

No

No

No

No

No

Yes

No

DT, data No cube

clustering, association

DT, clustering

DT

SN: Sequencing and Navigation Rules

[15]

MAS: Multi Agent System

Learning Management System Based on SCORM, Agents and Mining

301

considering that styles can change over time. A collection of agents was implemented to monitor student activities. Moreover, it allows the institution to base its work on SCORM and teachers to use different learning models according to the experience they have with each one of them. Differences between this work and related works are shown in Table 1.

3 Conceptual Model To understand all concepts involved in our proposal, we have divided the learning process into four stages. Each stage can be described in terms of roles, users, concepts and their relationships (see Fig.1): 1) A definition of models and learning styles, with questionnaires regarding each model that include questions and answers used by the LMS. Considering that teachers or institutions must have the freedom to choose and use the learning model they consider most appropriate, our work allows administrators and teachers to define the specific model and the questionnaires at the level of course templates or of specific courses. 2) A homogenous SCORM-based course template is defined. This template includes goals, resources and activities, AT, metadata, sequencing and navigation for all learners. Reutilization is an essential attribute promoted by this “homogenization” stage. 3) Using the course template, a teacher can customize a specific course promoting the concept of “free cathedra” that include, among others, activities/resources registry, learning model used by the course, a strategy (support/reject/leveling) depending on the student learning style. 4) Learners develop the course, beginning with a survey that reveals their learning styles according to the selected model, and then a customized AT or LIT is generated by the LMS to monitor and evaluate learning activities (learning based on learning styles). The Felder-Silverman and the CHAEA (Honey-Alonso Questionnaire about Learning Styles) models are considered in this paper. Felder and Silverman [5] propose a bipolar dimensions model depending on the kind of information perceived by the learner (Active/Reflective, Sensing/Intuitive, Visual/Verbal and Sequential/Global). CHAEA defines four learning styles (Activists, Reflectors, Theorists and Pragmatists - similar to Kolb [17]) according to how information is perceived and processed. Concepts related to learning models are shown at the top of Fig.1. To achieve content adaptation it was essential to extend the metadata of learning objects to include the relationship of each learning model with styles. For this reason, the concept of SCORM Metadata (plus) is shown in Fig. 1. Although this additional information is stored in a database, it is not present at the XML File of the learning object in order to maintain the original SCORM specification structure. Therefore, the AT is customized to be used for relevant learning objects inside a specific activity. Even though this is an improvement, it creates a work overload for teachers because more than one learning object is required for each activity.

302

C. Cobos et al. Learning Styles

t0

Learning Models

1

Questionnaires

Directors / Coordinators

Questions

Answers

Objectives “Homogenization”

Reutilization based on SCORM

Activity Types Indicators Resources (SCO’s)

SCORM Meta data (plus)

Activities Course Template

2

Aggregations Sequencing

Content structures

Directors / Coordinators

SCORM SN Roll Up

Personalized objectives

Reutilization based on SCORM

“Free cathedra” Personalized activities

3

Course Teachers

Personalized content structure

Personalized resources

Personalized aggregations Personalized SCORM SN

Preferences

“Learning based on learning styles”

Associated Learning Model

Obtained indicators Customized Course

4 Learners

Personalized sequencing Personalized Roll Up

% Learning Styles Answers of initial questionnaire

Base Learner Profile

New Learner Profile

Developed activities Learning Instructional template

SCORM Meta data (plus)

Agents Decision Engine

Decisions with C4.5 / ID3

t1

Fig. 1. AHS Conceptual Model. From top to bottom every concept involved in the customization stage for a learner’s AT is shown.

Of course, the system can operate with only one learning object per activity, but in that case there is no real content adaptation.

4 LMS Architecture The developed LMS (plus some standard LMS features such as registry, AT and learning object visualization, chat, forums, etc.) provides the work of a collection of agents cooperatively to support tasks like AT adaptation and learner activity monitoring.

Learning Management System Based on SCORM, Agents and Mining

303

The LMS has three different nodes: the content server and external repositories, the application server (Internet information server) and the Windows client application. The content server stores relational data, learning objects, etc., in a secure network server not accessible from the Internet. The application server stores web services in three layers (facade, business and services logic). Finally, the Windows client application establishes a communication with Web services through a user session with SOAP/XML. This application is based on smart-client architecture (see Fig. 2). Learners interact with the AT viewer, the learning objects viewer, the evaluating services and the anthropomorphic agents, among others. Content Server

SGBD

Windows Client Application

“Learning models” “Profile” “Learning objects” “…”

Files (SCO’s)

Application Server

Web Services Course Aggregation 1 Activity 1.1 Activity 1.2 Aggregation 2

Learning Objects Viewer

Aggregation 2.1

Internet

Activity 2.1.1 Aggregation 2.2 Activity 2.2.1 Activity 2.2.2

Learner

Aggregation 3

Evaluating Services Questionnaires

Facade

Business Logic

Service Logic

Secure LAN

HTTP/SOAP/XML Others …

Agents

Decision engine (ID3/Delivery)

Repositories

Fig. 2. LMS main components: Content server and repositories, application server and Windows client application

Agents are created in the Windows client application, but they have their logic in the application server (web services) and store all data in the content server. They are grouped according to behavior inside software packages. These software packages interact with the AHS using a communication bridge called “Agents Coordination”. Each package is described below. Interface Agents. This package contains two agents, the Interface Agent and the Anthropomorphic Agent. The main purpose of the Interface Agent is to perform the teacher’s role. It is responsible for communication tasks with learners. An Anthropomorphic Agent presents an animated anthropomorphic design inside a graphical user interface (GUI) to show that all messages are generated from other agents. Data Access Agents. This package contains two agents, the Profile Agent and the Learning Styles Agent. The Profile Agent is responsible for managing and storing the user’s identity in the content server according to content preferences, consumer settings and the user environment. Moreover, it recovers, modifies, and updates all learner data associated with a content. The Learning Styles Agent is responsible for storing the learner’s information and associated learning styles in the content server (e.g. in a neuro-linguistic programming model, a learner could be 50% visual, 30% auditory and 20% kinesthetic).

304

C. Cobos et al.

Adapter Agents. These agents are responsible for the creation of the AT and its presentation to the user through the Interface Agent inside the Content Structure Viewer (a content explorer) according to the learner’s specific, predominant learning style. These agents are capable of automatically recommending resources, sequences or navigation strategies according to a specific learning style. This package has two agents: the Adapter Agent That Supports Style and is responsible for finding resources related to a predominant learning style, and the Adapter Agent That Reject Style and is responsible for finding resources not related with a learner’s style. Monitoring Agents. These agents are responsible for continuously updating learner profiles by monitoring learning activities planned for a course; every learner event related to an activity will be stored in a profile. The main goal of this package is to systematize every experience had during a learning process and at the same time qualify the achievement of each proposed objective to set up learner strategies. Each agent proactively increases the learner knowledge and improves the performance of the system by continuously updating the learner profile using several techniques based on historical data and self evaluation. This package contains two agents, the Content Status Agent and the Indicator Monitoring Agent. The Content Status Agent creates part of the functionality of the SCORM execution environment. This agent is responsible for two things: a) updating the state of each node in the course content structure using a color code (e.g. blue for approved, green for completed, yellow for finished, red for revised and black for not revised), and b) linking itself to course content and changing its state based on resource visualization time indicators. The collection of Indicator Monitoring Agents is responsible for updating learner profiles based on activities performed by the learners. Each kind of activity has one or more indicators (comprehension level, number of times that a resource is being used, rest time during the activity and resource use time).

5 Agent’s Implementation The MAS-CommonKADS methodology [18] was used to analyze and design the system, since this methodology is based on knowledge engineering and aspects related to multi-agent systems are covered by it. These aspects are agent knowledge modeling, agent interaction modeling, and the integration of object oriented techniques. MAS-CommonKADS allows the inclusion of UML notation or AUML (Agent Unified Modeling Language) that is focused on the agent’s field [19]. On the other hand, it was essential to have a platform or API to create and control the agent’s life cycle because the implementation of the system was in VS.NET. A technological exploration was made of the tools used to develop multi-agent systems and only those tools which complied with the FIPA standards were considered. The tools found used Java as a platform making their interoperability with VS.NET technology difficult. Two alternatives were possible. The first was to achieve interoperability between Java and .NET platforms. (Bayo explains [20] how to build and instantiate agents in .NET through JADE.) The second was to develop an API built in .NET but according to FIPA standards. Initially, the interoperability between the JADE and .NET platforms was tried. However, some JADE services could not be used due to communication problems

Learning Management System Based on SCORM, Agents and Mining

305

between those technologies and this option was rejected. Finally, considering some ideas from JADE, a new .NET-based platform to control the agent’s life cycle was designed and implemented using C# as the programming language. The implementation of the anthropomorphic agent was developed with Microsoft Agent, a Visual Studio.NET library COM Object.

6 Data Mining Approach with ID3 and Delivering The agent architecture presented previously and a data mining approach are proposed to construct an adapted AT with an associated SN for each learner. A decision engine based on an ID3 implementation [8] uses a learner’s profile (which stores all user experience in the system) to create a unique decision tree (DT). This DT groups all the information about relevant experiences for further use, evaluating new experiences or making a decision (based on an algorithm called “Delivering”). Any kind of information can be stored in a learner profile: for example, gender, marital status, learning styles, hobbies, etc. Developing the decision engine required the use of a knowledge representation mechanism based on a relational model and an inference mechanism that uses DTs. Our learner profile includes the following learner attributes: learning model (LM), learning style (LS) and style level (SL, high or low depending on whether the style is predominant or not). The learner profile also includes System Attributes which control how contents are shown to the learner to promote or reject, for example, the learning style of students. These system attributes are the content structure tree (CST), the activity type (ATy) and an attribute called strategy, which has been created from the available learning styles to support or reject the style or leveling between one learning style and its opposite (see Table 2). Table 2. Learner and System Attributes From Learner attributes

Learner Attributes Learning Model Learning Style Style Level Content Structure Tree

System attributes

Activity Type Strategy

Values Felder, CHAEA Active, Reflective, Sensing, Intuitive, Visual, Verbal, Sequential, Global Low, Medium, High Choice, ChoiceExit, Flow, ForwardOnly Analysis and Reflection, Demo, Explanation, Enforce, Motivation, Evaluation, Review, Lecture, Simulation, Game Support, Reject, Leveling

The set of user experiences is registered by monitoring agents. There are three important events to monitor. The first is when the learner does all activities of an aggregation and presents the evaluation. In this case the system stores the experience and its attribute values for the learner and his/her evaluation results. (A Boolean value indicates if the learner approved or not.) The second event to monitor occurs when the learner decides to log off from the system without visualizing the entire number of

306

C. Cobos et al.

activities of an aggregation or presenting the evaluation. In this case the system will store the learner’s experience, its attribute values and a negative evaluation (indicating that the activities presented were not pleasant for the learner). The third event is when the learner asks the system to change the activities presented in an aggregation (asking the decision engine to generate a new strategy) and the experience report is negative. Those experiences will be presented as a tuple (data row) of the database as follows: S = (Learning-Model, Learning-Style, Style-Level, Content-Structure-Tree, Activity Type, Strategic, Result) S1 = (Felder, Global, Medium, Choice, Explanation, Support, Positive) Result {Positive, Negative}: if experience was appropriate or not.

(1)

Next, the ID3 algorithm uses attributes and the set of experiences for its training and generates the DT. ID3 was created by Quinlan in 1986 to make decisions using information gain criteria. Finally, “Delivering” generates a personalized AT for each learner. Delivering determines the application order of each strategy (from the most to the least recommended) based on the DT and strategies available in the course (See Fig. 3).

Implemented strategies for actual course Learners Delivering

Learner’s experience BD

ID3

Decision tree

Personalized AT Actual Learner

Learning and system attributes

Fig. 3. Components and inference mechanisms. The learners register their experiences. ID3 creates the DT. Finally the system generates the most appropriate AT for a specific learner.

The Delivering algorithm finds every route with a positive experience inside the DT and then takes the collection of values that produced a result to separate them into learner and system attributes. The learner attributes are used to compare the values of these same attributes inside the learner profile the system is processing at the time. The goal is to find the same combination and retrieve the collection values of the system that belong to a specific experience in the knowledge base. If there is not an exact combination of values, the case in which there is the greatest number of matched learner attributes is retrieved. If a specific experience for a learner is satisfactory with these system attribute values, there is a high probability that it will be repeated. Fig. 4 provides an example of a DT and shows a set of experiences.

Learning Management System Based on SCORM, Agents and Mining

307

Fig. 4. DT and Set of Experiences. A collection of positive and negative experiences registered by the system (right side) and the DT generated from this data collection (left side).

7 Test Results To evaluate the proposed model, a test with sixteen (16) learners (with an extensive set of experiences) was performed. The learners were students from the class “Introduction to Computer Science” in the first semester of the System Engineering degree course (computer science) at the University of Cauca. Two topics, “Algorithms” and “Meta-Algorithms”, were used. Each topic provided resources for all the learning styles supported. Two main aspects were evaluated on the test. First, the interaction with the LMS (see Fig. 5) and, second, a survey performed for each learner to evaluate that interaction. During the interaction, each learner completed Felder’s questionnaire (approximately 15 minutes) and then studied both topics mentioned above (30 minutes) using the adaptation support described previously. The general results of this test were the following: 1) the predominant style was visual; 2) 87% of the learners agreed with the content presentation; 3) 94% of learners considered that the AHS helps the learning process; and 4) 81% of learners considered that the content presentation related to “Algorithms” helped them pass the exam, while 80% also considered that content presentation related to “MetaAlgorithms” helped them pass the exam. The functionality test was performed to verify the adaptability process on content presentation, because every content structure presented to learners through the tool of the AHS module was generated/customized according the most predominant learning style of each learner. Of course, this preliminary test highlights the need to do a more formal experiment because the results are closer to a functionality test than a formal experiment.

308

C. Cobos et al.

Fig. 5. Learner’s Client Interface during test. 1) shows the course studied. 2) and 3) show the aggregations and activities with their different states according to colors previously explained in Section 4. 4) shows the learning object studied. 5) shows the Anthropomorphic Agent.

8 Conclusions and Future Work In order to model and develop our LMS, an agent-based approach was essential (robustness, scalability and simpler programming). The final product supports several learning models, provides a decision making module that uses the ID3 algorithm and is based on SCORM SN (SCORM Sequencing and Navigation Specification). The results of the tests are appropriate because the system really adapts contents, presentation and navigation to the user needs. In a near future, we plan to include more attributes in the learner’s profile, e.g. gender, learning motivation, media preference, hobbies and social status. In addition, a more controlled and formalized experiment will be designed to determine the effects derived from the use of the LMS in the learning process. Finally, the implementation of the decision engine using an EC4.5 [21], C4.5 or C5.0 [22] algorithm will manipulate discrete attribute values and in this way eliminate the conversion between a percentage value to a high-medium-low value.

References 1. Paulsen, M.F.: Online Education and Learning Management Systems. Global E-learning in a Scandinavian Perspective. Bekkestua: NKI Forlaget, p. 337 (2003), ISBN: 82-5625894-2 (printed version), http://www.studymentor.com/studymentor 2. SCORM. Sharable Content Object Reference Model (SCORM) (2004) (retrieved September 22, 2005), http://www.adlnet.org 3. Sharable Content Object Reference Model (SCORM) Sequencing and Navigation (SN) (retrieved September 22, 2005), http://www.adlnet.org/scorm/history/2004/documents.cfm

Learning Management System Based on SCORM, Agents and Mining

309

4. Chang, H.-P., Wang, C.-C., Jan, K.H., Shih, T.K.: SCORM sequencing testing for sequencing control mode. Advanced Information Networking and Applications. In: 20th International Conference on AINA 2006, vol. 2 (April 18-20, 2006), DOI 10.1109/AINA. 2006.295 5. Felder, M.R., Silverman, L.: Learning and Teaching Styles in Engineering Education. Engineering Education 78(7), 674–681 (1988) 6. Brusilovsky, P., Maybury, M.: From Adaptive Hypermedia to the adaptive web. Communications of the ACM 45(5) (2002) 7. Oppermann, R., Rashev, R., Kinshuk: Adaptability and Adaptivity in Learning Systems. In: Behrooz, A. (ed.) Knowledge Transfer, pAce, London, vol. II, pp. 173–179 (1997) ISBN 1-900427- 015-X 8. Quinlan, J.R.: C4.5: Programs for Machine Learning. Morgan Kaufmann, San Francisco (1993) 9. Shang, Y., Shi, H.C., Chen, S.S.: An intelligent distributed environment for active learning. ACM Journal of Educational Resources in Computing 1(2), 1–17 (2001) 10. Trantafillou, E., Poportsis, A., Demetriadis, S.: The design and the formative evaluation of an adaptive educational system based on cognitive styles. Computers & Education 41, 87– 103 (2003) 11. Gilbert, J.E., Han, C.Y.: Adapting instruction in search of a significant difference. Journal of Network and Computer Application 22, 149–160 (1999) 12. Weber, G., Specht, M.: User modeling and adaptive navigation support in www based tutoring systems. In: Proceedings of User Modeling 1997, pp. 289–300 (1997) 13. Chen, G.D., Liu, C.C., Ou, K.L., Liu, B.J.: Discovering decision knowledge from web log portfolio for managing classroom processes by applying decision tree and data cube technology. Journal of Educational Computing Research 23(3), 305–332 (2000) 14. Peña, C.I., Marzo, J.L., De La Rosa, J.L.L.: Intelligent Agents in a Teaching and Learning Environment on the Web. In: ICALT2002. 2nd IEEE International Conference on Advanced Learning Technologies, Kazan (Russia), pp. 21–27 (September 9-12, 2002) 15. Wang, F.H., Shao, H.M.: Effective personalized recommendation based on time-framed navigation clustering and association mining. Expert Systems with Applications 27, 365– 377 (2004) 16. Su, J.-M., Tseng, S.-S., Wang, W., Weng, J.-F., Yang, J.T.D., Tsai, W.-N.: Learning Portfolio Analysis and Mining for SCORM Compliant Environment. Educational Technology & Society 9(1), 262–275 (2006) 17. Kolb, D.: Experiential learning: experience as the source of learning and development. Prentice Hall, Englewood Cliffs, NJ (1984) 18. Iglesias, C.A., Gariji, M., Centeno-González, J., Velasco, J.R.: Analysis and Design of Multiagent Systems Using MAS-Common KADS. In: Rao, A., Singh, M.P., Wooldridge, M.J. (eds.) ATAL 1997. LNCS, vol. 1365, pp. 313–327. Springer, Heidelberg (1998) 19. Odell, J., Muller, P.J., Bauer, B.: Agent UML: A Formalism for Specifying MultiAgent Interaction. In: Ciancarini, P., Wooldridge, M.J. (eds.) AOSE 2000. LNCS, vol. 1957, pp. 91–103. Springer, Heidelberg (2001) 20. Bayo, A.M.: Running Jade over .NET (2003) 21. Ruggieri, S.: Efficient C4.5. IEEE Transactions on Knowledge and Data Engineering 14(2), 438–444 (2002) 22. Larose, D.T.: Discovering Knowledge in Data: An Introduction to Data Mining. John Wiley & Sons, Incorporated, Chichester (2005)

Suggest Documents