ijns05112012

of 6

Please download to get full document.

View again

All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
PDF
6 pages
0 downs
1 views
Share
Description
Jaber EL BOUHDIDI et al., International Journal of Networks and Systems, 1(1), August – September 2012, 26 - 31 26 @ 2012, IJNS All Rights Reserved  ABSTRACT Several researches in the field of education have shown that taking into account learning styles has drastically improved the quality of teaching / learning. The adaptation of the course into the profiles and preferences of learners requires the collection of more informa
Tags
Transcript
  Jaber EL BOUHDIDI   et al ., International Journal of Networks and Systems, 1(1), August – September 2012, 26 - 3126@ 2012, IJNS All Rights Reserved      ABSTRACT Several researches in the field of education haveshown that taking into account learning styles hasdrastically improved the quality of teaching / learning.The adaptation of the course into the profiles andpreferences of learners requires the collection of moreinformation on learners, learning styles andeducational resources.To identify the learning style of each learner, thearchitecture is designed to require the learner to passthe test of Felder and Silverman in his first connection.This test provides information about the preferences of learning styles of the learner.Our contribution in this paper consists of an adaptiveapproach based on the semantic web and Bayesiannetworks (BN), to provide learners with personalizedcourses according to their profiles and learningobjectives. In addition, the system allows to make adiagnosis and classification of errors made by learnersto generate relevant remedial course. Indeed, thismodel allows learners, teachers and instructionaldesigners to work with software agents toautomatically and effectively build custom routesoriented educational goals. Key words : Adaptive Learning Paths, Bayesian Network,Learning Object, Multi-agent System. 1.   INTRODUCTION The adaptation of learning paths to learners' profiles requiresthe collection of more information about learners and learningresources. This article aims to develop a model of E-Learningto generate personalized learning paths for a group of learnerswith similar characteristics and common goals. The proposedmodel is designed based on a multi-agent paradigm, thesemantic web and Bayesian networks. A phase which isessential to ensure that adaptation of course, is theclassification of learners. It is done by the system based on amulti-agent architecture [1]. These agents are in constantcommunication and cooperation between them in order topromote the recognition of the learner profile (preferences,learning style, knowledge and skills, etc..) And assign it to aclass. The classification is imposed on the learner during hisfirst connection and can be requested by a student to changeits class membership already registered who wants tomanually update your profile or required by the system in casethe learner performance is modest.To this classification, we chose the naïve Bayesian classifiersbecause they have demonstrated an efficiency more thanenough in many real and complex situations. The advantageof naive Bayes classifier is that it requires relatively littletraining data to estimate the parameters needed to classify.In other words, we modeled three ontologies: ontologylearning resources to represent teaching materials, ontologylearning objectives according to Bloom's taxonomy andontology to model the learner's profile.In addition, we kindly use the classification results at multiplelevels in addition to creating customized learning paths willalso create routes remediation for learners who have not beensuccessful goal. For this, the system will make a diagnosis andclassification errors made by learners to generate these routesthat attempt to correct errors and fill gaps learners. Thismethod will allow us to reuse already created remedialcourses, optimizing their creation time and to target thedifficulties learners. 2. REPRESENTATION OF RESOURCES   In our approach, we use ontologies for describing thefeatures of domains. The contribution of ontologies forunderstanding, sharing and integration of information isdemonstrated. Indeed, research and practice in this area beginto bear fruit, especially for the Semantic Web. Threeontologies were designed: Ontology of pedagogical resources,Objectives Ontology, Ontology of learners’ profiles.Jaber EL BOUHDIDI, Mohamed GHAILANI, Abdelhadi FENNAN 1 Laboratory LIST FST-Tangier, Morocco, jaber.f15@gmail.com 2 Laboratory LIST FST-Tangier, Morocco, ghalamed@gmail.com 3 Laboratory LIST FST-Tangier, Morocco, afennan@gmail.com Architecture for the adaptation of learning paths based on ontologies andBayesian networks   ISSN 2319 - 5975 Volume 1, No.1, July August- September 2012International Journal of Networks and SystemsAvailable Online at http://warse.org/pdfs/ijns05112012.pdf     Jaber EL BOUHDIDI   et al ., International Journal of Networks and Systems, 1(1), August – September 2012, 26 - 3127@ 2012, IJNS All Rights Reserved   2.1 Learning Object Learning objects (or Unit of Learning) are smaller units of learning and is currently at the heart of many applications of instructional design. Current work on learning objectsinterested in the standardization of these based on metadatadescribing their content to ensure case pedagogicalproductions in what is called the education market. Generally,this standardization is based on different research directionsor describe learning objects as entities that the system has andwhich manipulations are based on the metadata specification,either towards the educational modeling languages forrepresent Hypermedia Units of Learning. Three approacheshave emerged and led successively on each of the standards orproposed standards: LOM, SCORM and IMS LearningDesign. The term learning object emerged in the mid 1990sin international consortia such as IMS and ARIADNE - whichled him to propose a standard. In this paper we use the LOM[2] standard for the representation of educational resourcesLOM (Learning Object Metadata) in the early 2000s. The goalis then profitable production and develop reuse (economicperspective). Several standardization of metadata foreducational resources have been conducted, [4, 5]. 2.2 Structure of the Training Modules  Our architecture is based on the pedagogy by goals tostructure the material to teach (i.e. the learning module), weuse a three-level hierarchy of educational objectives asdefined in [6, 7]:1.   The General Objectives or abstract (GO);2.   The Specific Objectives or composite (SO);3.   The Operational Objectives or atomic (OO);To classify these objectives, we opted for the taxonomyof cognitive domain by Benjamin BLOOM, who is the fatherof the first hierarchical classification of educationalobjectives. The taxonomy of educational objectives BLOOM[8],is composed of six levels, including: knowledge,comprehension, application, analysis, synthesis andevaluation. For each class, there is a set of verbs that can beused to express the objectives of educational services.This hierarchy has allowed us to consider three levels of abstraction module of instruction:1.   Parts (meeting the General Objectives);2.   Chapters (that meets the Specific Objectives);3.   Hypermedia Learning Units (Object Learning)(meeting the operational objectives).These are transfer credits evaluated. The system, then,organizes the process of education around these componentshypermedia (the L.O). The LOs are supposed to receive, byinstantiation, all kinds of domain knowledge in all forms of media permitted by HTML (text, image, sound, video, script,applet), Figure 1 shows the structure of a module into simpleelements.The sequence of learning objectives (LOs) by the system ismade on the basis of a network of pre-requisites proposed bythe author of the teaching module. A prerequisite link betweentwo objectives LO1 and LO2 (from LO1 to LO2) defines on theone hand a precedence desired by the author between the twoobjectives, proposing that learning the second objective cannotbe completed until LO2 achievement (or success) of the firstgoal LO1, on the other hand, a link indicative of progression ora remediation of a potential link. This latter feature means thatthe system can choose a LO that is a pre-requisite to a LO onwhich the learner has failed in order to offer him a contributionof knowledge that relates to the LO prerequisites. 3. LEARNER MODEL   The learner model is a model for representing the informationof the student come into play when building a suitablelearning path. It also allows the system to adapt to the learnerwho interacts with him. That is to say, it has the knowledge tounderstand and use what the learner already knows.This can be achieved through knowledge of the learnerprofile. This profile must integrate the knowledge of the learner on the field, but also can add features to thelearner as its educational objectives, preferences ....etc.To design the model of learning there are two majorstandards that can be adopted. This is the PAPI (IEEE / PAPI Public and Private Information for Learners - Information on public and private learners ) and IMSLIP (IMS / LIP Instructional Management SystemsGlobal Learning Consortium for Learner InformationPackage ). Both standards specify several categories of information about the learner. PEDAGOGY BY GOALS Structuring Module OPERATIONAL OBJECTIVESSPECIFIC OBJECTIVESGENERAL OBJECTIVES B L  O OM’   S T  ax on om  Part 1Part N……...Chapter 1 ……..Chapter N  L.O   N   L.O   1   ……..   Figure 1: Hierarchical representation of a module  Jaber EL BOUHDIDI   et al ., International Journal of Networks and Systems, 1(1), August – September 2012, 26 - 3128@ 2012, IJNS All Rights Reserved  In this paper we adopt the standard IMS-LIP[3,8, 9,10]it is based on a data model that describes the basiccategories to record and manage the academicbackground, training objectives, and outcomes of learners. These categories are described in Figure 2. Figure 2: Hierarchy of concepts of the IMS- LearnerInformation Package (LIP). The elements of the LIP specification are: theidentification that represents the data on thedemographic and biographical studies. Goalrepresenting the study objectives and aspirations of thelearner. The QCL (Qualifications, Certificates &Licenses) which as its name implies is thequalifications, certifications and permits granted bycompetent authorities to the learner.Activity that represents all learning activitiesregardless of their state of completion, includingformal and informal education, training and work experience.The element transcript is a record for a summary of activities based on academic achievement. The interestelement that represents the information on recreationand activities outside of work and school. The elementthat represents competency skills, knowledge andskills acquired in the cognitive, affective, orpsychomotor. The affiliation is part of the studentmemberships in professional organizations. Theelement that characterizes the accessibility informationaccessibility to the learner as defined by thepossibilities of language, disabilities, acceptability andpreference studies including cognitive preferences (c tod learning style), physical preferences (c to d apreference for large page), and technologicalpreferences (c to d preference for a particular computerplatform). The security element key is the set of passwords and security keys assigned to the learner fortransactions with the systems and information servicesfor learners. Finally the element relationship for allrelations between components of the nucleus. 4.   THE LEARNING STYLE MODEL Learning Style (LS) can be defined as the way a personcollects, processes and organizes information. Among thedifferent proposals for modeling LS, we choose the FSLSsince it is one of the more successful models and has beenimplemented in many e-learning systems. FSLS classifiesstudents in four dimensions [12]:  Active / Reflective (Processing). Active people considerhaving understood a piece of information only if they havediscussed it, applied it or tried to explain it to other people.Reflexive people, on the other hand, prefer reflecting aboutthe issue before assuming any practical posture. Sensing / Intuitive (Perception). Sensing people are meant tolearn from tasks related to problems and facts that could besolved by well-behaved methods, with no surprises orunexpected effects. Besides, this style usually refers tostudents that are fond of details and very good memorizers of facts and practical applications. Conversely, intuitive studentsare meant to discover alternate possibilities and relationshipsby themselves, working with abstractions and formula, whichallows them to understand new concepts and to quickly andinnovatively perform new tasks. Visual / Verbal (Input). Visual-driven people find nodifficulties in interpreting, for an example, pictures,diagrams, timelines or movies. Distinctly, verbal students’personal learning processes are driven by written or spokenexplanation. Sequential / Global (Understanding). Sequential peoplestructure their learning process by logically, successivelychained steps, each one of them related to the search forsolutions. On the other hand, global students learningprocesses are distinguished by random jumps: they often areable to solve a complex problem, although they do not knowhow they arrived at the solution.Felder and Silverman proposed a psychometric instrument,the Index of Learning Style Questionnaire (ILSQ), thatclassifies the preferences for one or the other category as mild,moderate or strong. In the majority of traditional AEHS that  Jaber EL BOUHDIDI   et al ., International Journal of Networks and Systems, 1(1), August – September 2012, 26 - 3129@ 2012, IJNS All Rights Reserved  make use of a learning style model for adaptive purposes, theassumptions about the student’s learning style are usuallyacquired by a psychometric instrument like ILSQ.Nevertheless, the use of such a test has some drawbacks.First, students tend to choose answers arbitrarily. Second, it isreally difficult to design tests capable of exactly measuring“how people learn”. Therefore, the information gatheredtrough these instruments encloses some grade of uncertainty.Moreover, this information, as a rule, is no longer updated inthe light of new evidences from the student’s interactions withthe system. An alternative approach that uses a BayesianNetwork (BN) to model the student’s LS, instead of acquiringit by a psychometric test. Using a BN as a LS model allowsthat observations about the user’s behaviour can be used todiscover each user’s LS automatically using the inferencemechanisms. 5.   THECLASSIFICATION OF LEARNERS Several ideas have emerged over the years on how to obtainrelevant results for classification, so there are differentapproaches that can be used to a degree such as: clustering,Bayesian Networks, Neural Networks, Support VectorMachines (SVM), etc. In this paper, we used Bayesiannetworks to classify the learners, the classification of studentsis done in two phases: first when a student logs into the systemfor the first time, the system requests information on theirprofile. These will be stored in a file OWL, which will be usedby the classifier agent. It affects the learner at first level of agiven class based on cognitive preferences, preferences,physical and technological preferences. Then, when a studentmakes a goal to apply for a learning path, and if the course hasprerequisites, the system generates a pre-test to verify theseprerequisites, the results of this test are used to assign thelearner at a particular level of the class itself. In addition, atthe end of each course, the learner passes a test to pass thegoal. The system decides the outcome of the following tests tocreate a remedial course or not. We will explain later in detailthe principle of creating a course of remediation whileclassifying the errors made by the learner. 5.1 Bayesian Classifiers Bayesian classifiers are statistical classifiers. Theycan predict classmembership probabilities, such as theprobability that a given sample belongs to a particularclass.Bayesian classifier is based on Bayes’s theorem.Naive Bayesian classifiers assume that the effect of anattribute value on a given class is independent of thevalues of the other attributes. This assumption is calledclass conditional Independence. It is made to simplifythe computation involved and, in this sense, isconsidered ”naive”.Depending on the nature of each probabilisticmodel, the naive Bayesian classifier can be trainedeffectively in the context of supervised learning. Inmany practical applications, parameter estimation fornaive Bayesian models based on maximum likelihood.In other words, it is possible to work with the naiveBayesian model without worrying about Bayesianprobability or using Bayesian methods.The naive Bayesian classifier showed an efficiencymore than adequate in many complex real situations.The advantage of the naive Bayesian classifier is that itrequires relatively little training data to estimate theparameters required for classification, ie means andvariances of different variables. Indeed, the assumptionof independent variables can be satisfied with thevariance of each of them for each class, without havingto calculate covariance matrix. The probabilistic modelfor a classifier is the conditional model :  p(C|F  1  ,F  2  ,….,F  n ) where C is a variable dependent class or classes whoseinstances are few, determined by several characteristicvariables F  1  ,…..,F  n  When the number of features n is large, or when these featurescan take many values, this model based on probability tablesis impossible. Therefore, we derive to be more easily soluble.Our classification algorithm is based on the NB approach. Thestandard Bayes rule is defined as follows:  max {  (  |  )}=  (  |  ) ∗ (  )  (  )   (1)  Where;P(C n )= the prior probability of category n,W = the new profile to be classified,P(w|C n )= the conditional probability of test profile, givencategory n.The P(w) can be disregarded, because it has the same valueregardless of the category for which the calculation is carriedout, and as such it will scale the end probabilities by the exactsame amount, thus making no difference to the overallcalculation. Also, the results of this calculation are going to beused in comparison with each other, rather than as stand-aloneprobabilities, thus calculating P(w) would be unnecessaryeffort. The Bayes Theorem in (eq. 1) [13] is thereforesimplified to: argmax{  (  |  )} ∞ (  |  ) ∗ (  )  
Advertisements
Similar documents
Advertisements
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks