Abstract : The objective of this thesis is to create a new system that can create adaptive and intelligent educational system according learning styles and the state of knowledge of each student. This system is called IWEBISE (Intelligent WEb-Based Interactive System for Education). It is the result of the coupling across five key areas: knowledge engineering, human-computer interaction, cognitive psychology, artificial intelligence and psychology. IWEBISE consists of five parts:
Student Model: This section details how the learning style of each student is modeled using the Felder-Silverman model, which depends on many parameters: Number of examples, number of exercises, examples before or after content and exercises before or after content. Students knowledge is expressed in the system by using the overlay model which considers it a part of the knowledge domain. The IWEBISE also uses an opened student model, which allows students change by themselves their knowledge status related to each concept in domain knowledge; this permits them to study with wide steps without feeling blocked during their learning process. The student model is composed of two parts: Static part: It keeps student's personal information. Dynamic part: It maintains the record of the students' understanding as the course progresses on the basis of their responses. The personalization of navigation through the course content depends on several parameters taken from the interaction of student with the system. These parameters are: number of correct answers (NCA), number of incorrect answers (NICA), time spent to solve a question (TSSQ), time spent to reading or interacting with a specific concept (TSR) and number of attempts to answer a question (NAAQ). Once a student passes a pre-test session, the dynamic part is initiated using these parameters. Using these parameters six methods are employed to symbolize student's knowledge status in six levels (Excellent, very good, good, rather good, weak and very weak) with the objective to determine the best one to be used later within IWEBISE. These methods are: FBAM, ART2, Fuzzy-ART2, HMM and NN/HMM. All previous algorithms performance is evaluated by employing F-measure metric to determine the best one to be used later in our new system. Results show that Fuzzy-ART2 gives best categorization quality (0.281 as it is depicted in the following table), which is considered a very important factor to assure that an appropriate course map is displayed to the student according to his/her knowledge status. This pushes him/her to finish the course completely without feeling boring and lost in it.
Tutor Model: This section explains in details how to model different teachers' strategies in presenting course content to students. This is done by having them in a table consisted of nine fields which allow to store colors utilized to represent students' knowledge status in the course map and the possibility of showing or hiding a learning concept. This model also focuses on a prediction algorithm to foresee the next concepts might be visited by students. The prediction process is achieved by following three phases: Initialization phase: For each student a HMM (λ) is built based on his/her previous concept access sequences. Adjustment phase: Given a new observed sequence and a HMM (λ), the Baum-Welch algorithm is used to adjust the initialized HMM and to maximize the new observed sequence. Prediction phase: The Forward Algorithm is applied to determine the probability distribution of each concept (state) in the course. The highest value represents the next concept will be visited by the student. Finally, the students' actions prediction accuracy is measured with the recall (sensitivity) and precision measures. The sensitivity is defined as the number of correctly predicted concepts (true positives) divided by the number of annotated concepts (actual positives). The precision is the percentage of positive predictions that are correct.
Domain Model: The course content is organized into a concept network to represent learning objectives. A learning objective (LO) concerns several concepts which are classified in: main concepts (MC), prerequisite concepts (PC) and sub-Concepts (SC). Each internal node in the network represents a concept, and external nodes in the lowest level symbolize several types of educational units (EU), which are in the form of interactive flash multimedia files, images, videos, texts, exercises, examples and tests. Three methods are used to model course content with the objective to choose later the best one of them: In the first method the domain knowledge is conceived and modeled using a hierarchical BAM neural network. The first BAM-1 is to associate learning objectives with concepts and while the second BAM-2 is utilized to assign educational units to each concept. The output layer of BAM-1 is the same input layer of BAM-2 which can be seen as an intermediate layer of the whole architecture. Number of nodes of the input, middle and output layers represent number of learning objectives, concepts and educational units. In the second method a relational database is used to represent domain knowledge. It consist of eleven tables (Main Category, Subcategories, subjects, Learning objectives, Prerequisite-learning objectives, concepts, Prerequisite-concepts, attach a sub-concepts with concepts, concept content and test items). A Document Type Definition (DTD) file is constructed in the third method to determine a set of rules to define and describe the organization of knowledge within an XML file. Relational database is selected to be used within our new system IWEBISE due to the following reasons: Some course designers prefer to have their course contents confidential and protected. XML could not treat huge course contents and all kind of data such as images and video. Users Interfaces: This part uses some windows used to facilitate the interaction among different users of IWEBISE. These windows are classified in four levels: Administrator Interface: It permits administrators to create course category, subcategory, users management and subscription process. Designer Interface: It permits course designers to manage learning objectives, learning concepts, subconcepts, concept content and tests questions. It also permits them to package the whole course under SCORM standard. Tutor Interface: Teachers can mange their teaching strategies rules and trace students and give them the appropriate advises when they deviate from the final goal of the course. Student Interface: Students can complete their learning process by using pretest, post-test, “Index Learning Style” questionnaire, glossary, chat and forum windows.
Adaptation Engine: This parts presents the two adaptation algorithms used in tailoring the two following points to student: Concept content concerning students' learning style depending on Felder and Silverman model. Learning concept map according to students' knowledge status. In addition chapter IX shows: How different adaptation technologies are used in IWEBISE such as: Adaptive Presentation, Adaptive Navigation and curriculum sequencing. How constructive theory is applied and used by the intelligent tutor.
The originality of this thesis is based on: 1.The discover and the use of a new neural network architecture called hierarchical HBAM for modeling domain knowledge of a course. This new network can be employed in many other fields such as: pattern recognition. 2.The use of a new hybrid algorithm using a neural network (Fuzzy-ART2) and a statistical method (HMM) in modeling student's knowledge. 3.The comparison of many machine learning algorithms such as: FBAM, ART2, Fuzzy-ART2 and a hybrid structure Fuzzy-ART2/HMM, which are used for categorizing students' thinking and reasoning into six levels; very weak, weak, fair, good, very good and excellent. 4.The use of HMM to predict the next concept, based on the history of concepts, visited by a certain student navigating within the course; 5.Discovering Higher Institute of Languages (University of Aleppo) students' learning styles in learning a language. This is done by using Felder and Silverman model. 6.The comparison of the new system IWEBISE with others adaptive and intelligent educational systems. 7.The ability of the new system IWEBISE to export and package courses according to SCORM standard with the objective to reuse them within others educational platforms such as: MOODLE. 8.The construction and implementation of a new intelligent and adaptive web-based educational system for teaching English grammar.