Introduction

The role of the Pedagogical Agent[s] in DesignFirst-ITS is that of tutor who provides support to the student when he is experiencing difficulty grasping or applying a concept. This support is tailored to a particular student based on his individual profile that is maintained by the Student Model. The role of the Pedagogical Agent is discussed in detail in following sections.

1. Eclipse IDE:

The student’s interaction begins with the ITS when he begins to work on his solution in the UML interface of Eclipse IDE. The Expert Evaluator collects data as the student’s work progresses, periodically sending it to the Student Model. When the student makes an error, the incorrect action, the correct action and related concept are sent to the Student Model. The Student Model uses this information to update the student’s knowledge state and to come up with possible reasons and associated probability for student making the error. The Pedagogical Agent receives the error and diagnosis from the Student Model and decides on the appropriate feedback based on the history of instruction already offered.

Once the student makes an error, the Pedagogical Agent has to make the following decisions.

- What level of feedback and tutoring does the student need?
- What strategy will work best for the given student?

The answer to these two questions depends on the student profile that is maintained by the Student Model. To answer the first questions, the pedagogical model will look at the results of the diagnosis performed by the Student Model to see what concepts the student doesn’t seem to understand. Based on the understanding level of each concept, the Pedagogical Agent will come up with the feedback. The Student Model has to come up with an algorithm to assign understanding levels for each concept for the given student. If the level of understanding for a concept is low, then the Pedagogical Agent will come up with a detailed lesson plan, and if the understanding is medium, then the lesson will be brief and if the understanding of a concept is relatively high, then the Pedagogical Agent might just give a very brief overview of the concept along with an example.

To answer the second question, the Pedagogical Agent once again has to look at the Student Model to determine which strategy is best for the student.

2. CIMEL multimedia:

CIMEL multimedia is one of the two interfaces through which students can interact with DesignFirst-ITS. The multimedia is an interactive tutoring facility that will educate the students about the concepts of object oriented design and programming. During this tutorial, the students will be presented with exercises covering each section and an end of chapter quiz that will test his/her knowledge of the material.

When the student is going through the tutorial in CIMEL, all information about the session will be saved. Especially the results of student’s performance on the exercises and quiz will be interpreted by the Student Model. If the student profile in the Student Model indicates that the student did not perform well on the exercises or the chapter quiz, the Pedagogical Agent will become active and will attempt to interact with the student.

It will essentially tell the student that according to his profile, he seemed to have difficulty with certain concepts because he was not successfully able to answer or had difficulty answering certain questions in the exercises/quizzes that pertain to these concepts. It will offer the student help in trying to clarify the concepts. In order for the Pedagogical Agent to be able to provide assistance to the student, it requires the following

- All object oriented design and programming concepts from CIMEL multimedia to be represented in the curriculum information network.
- Each section in CIMEL must be indexed by the concept.
- All questions in CIMEL exercises and quizzes must be tied to specific concepts.
- All concepts have explanations attached to them through the feedback network.
- The student profile should indicate whether the student is a beginner, intermediate or advanced in object oriented design and programming. It should also have information that specifies which concepts the student knows well and which ones he does not. This concept comprehension must take into account the student’s performance on prerequisite concepts.
- Pedagogical Agent also needs to know lesson plan for the chapter in CIMEL and needs to know if the student followed the entire lesson plan or skipped some sections in the middle.
- It should keep a stack of concepts covered so far including the current concept that should be on the top.

Once the Pedagogical Agent determines that it is time to intervene, it looks in the Student Model to see what concept the student seems to be having difficulty with and what concepts he has covered so far in the lesson. The CIMEL multimedia interface will be implemented in the second phase of DesignFirst-ITS.

Categorizing student Learning Style

The Pedagogical Agent feedback and tutoring strategies will be personalized for each student based on their individual learning styles. The Felder-Silverman learning style model will be used to categorize learning style of each student. This model categorizes the students into five different categories but only four categories will be relevant for Pedagogical Agent in this system.

Each learning category contains a pair of styles that represent opposites for that category. The student can either fall right is the middle of the scale not favoring any of the two learning styles or he can fall towards slightly left or right of the midpoint favoring one learning styles over another or he can fall in the extreme right or left of the midpoint indicating a strong preference for one of the learning styles. The four learning categories are
a. Sensing Learner / Intuitive Learner
b. Visual Learner / Verbal Learner
c. Active Learner / Reflective Learner
d. Sequential Learner / Global learner

ILS (index of learning styles) will be used to categorize learning styles of each individual student. This is a form containing 44 questions that the users of the ITS will have to fill out the first time they use the ITS. The student responses from the ILS will be used to categorize a student’s learning style.

Teaching Strategies

The system will use strategies that match with the student’s learning styles. The higher level strategies will be the traditional tutoring model which will only have one Pedagogical Agent acting as the tutor and the co-learner model which will have two Pedagogical Agents, one acting as a tutor and the other agent in the role of a virtual student. In addition to the higher level strategies, the feedback/tutoring will also be customized by low level teaching strategies. For example, if a student is active and visual learner, then a multi-agent system with a tutor and co-learner will be used and the feedback content will mainly use diagrams and other visual tools. If the student is active and verbal learner, then multi-agent strategy will be used and the feedback content will mainly consist of written explanations.

Components

The Pedagogical Agent system will have the following components
a. FIN – Feedback Information Network
b. PSF – Problem Specific Feedback
c. AGNTS - One or more Pedagogical Agents
d. LSC – learning style categorizer
e. SFA - Strategy & feedback assembler
f. UI - User Interface

The Feedback Information Network (FIN) contains the feedback for each concept in the curriculum information network (CIN). The feedback is broken into different levels and different types. The levels indicate if the feedback is basic or advanced while the types refer to type of learning style it is appropriate for. The FIN is created in the implementation phase and does not change unless more information is added by the designer or instructor. The FIN only contains problem independent feedback that concentrates on object oriented design and programming concepts. The strategy and feedback assembler will assemble different types and levels of advice for a student based on their learning style.

The Problem Specific Feedback (PSF) component will contain the feedback that is associated with a specific problem. This feedback is entered by the instructor during a new problem entry phase that is handled by the Expert Evaluator. This feedback will be used in conjunction with the domain specific feedback from the CIN to design the feedback/tutoring session.

The first component that interacts with the student is the LSC which becomes invoked as soon as the student logs into the ITS. If it is the first time that the student is logging on, then the LSC presents the student with the “index of learning styles” questionnaire. Once the student completes the questionnaire, the student is able to work on his assignment in eclipse environment. In the background, LSC uses the completed questionnaire to categorize student’s learning style. Once LSC has categorized the student’s learning style, it passes this information to the Student Model so that it can become a permanent part of the student profile.

The learning style of a student should stay the same unless the student responds better to advice that is appropriate for a different type of a learning style. For example if the student’s initial learning style is determined to be visual but the student seems to understand the concepts better when presented with words, then LSC might be invoked to update the student’s learning style. In this case, the LSC will use the previously gathered student learning style information and the feedback effectiveness information to decide if the student’s learning style should be automatically updated. Feedback effectiveness information will be generated based on how well the student does on the homework with a concept on which he received feedback / tutoring help.

The fourth component of the pedagogical system is the strategy and feedback assembler (SFA) . This component uses the learning style preference of the student from Student Model as well as student’s understanding level of various concepts along with the information from the FIN and PSA to assemble a tutoring session that will be presented to the student. Depending upon the learning style preference of the student, the feedback will have different components and different level. Once the feedback and strategy assembly is complete, it will be passed on to the AGNTS which will then interact with the user.

The AGNTS will not only deliver the feedback to the student but will also keep history of the feedback provided to the student. This history will contain the strategy used, the actual feedback, the concepts the feedback was intended to help with and number of times feedback was required for the concepts in question. This historical data will be used by the strategy and feedback assembler for future feedback. The feedback and strategy will be marked successful if the student shows understanding of the concepts after he receives the feedback.

The user interface will consist of a persona representing the ITS tutor agent along with the feedback that has to be delivered. The user interface will slightly differ based on the student’s learning style. For example if the student is an active learner, then the interface might have a second agent persona in the form of a virtual student.

Pedagogical Agent Data Requirements

The Pedagogical Agent will interact directly with the Student Model when the Student Model indicates that the student has made a mistake or seems to be struggling. The data that the Pedagogical Agent needs in order to plan and deliver the feedback is as follows.

a. Student knowledge status – beginner, intermediate or advanced.
b. Student comprehension level of relevant concept in the curriculum.
c. Current problem –
- Has student made a specific mistake? If yes, the specific error code/ reason and related concepts
- If student has not made a specific mistake but seems to be stuck or struggling at a certain point, then the concept that student is stuck on or is struggling with
- if student has pressed the help button, then the concept that student needs help with
- Model of the homework problem that the student is struggling with
d. Feedback history – what feedback was already provided to the student? How many times the feedback provided for the same concept. Has the student already gone through CIMEL multimedia? How did the student perform on CIMEL exercise and quizzes?

The Student Model will send a message to the Pedagogical Agent on three different occasions. The first occasion is when the Student Model receives an error packet from the Expert Evaluator indicating that the student has made a specific error. In this case, the Student Model will send the specific error and concept that the student has problem with along with a list of possible reasons and associated probability. The Pedagogical Agent will use this information to decide what tutoring/feedback the student needs. In some instances, the Pedagogical Agent might use the student learning style profile to eliminate certain reasons from the list sent by the student model.

The second occasion that Student Model will approach the Pedagogical Model is when the Student Model receives the “Eval” packets from the Expert Evaluator (EE) indicating that the student has pressed the “Evaluate” button. If the EE finds errors in student’s solution, it will send error packet to the Student Model which initiates the flow described in the previous paragraph. Otherwise, if the EE successfully completes evaluation of student’s work without error, the Student Model will notify the Pedagogical Agent so that positive feedback can be provided.

The third occasion is when the student presses the “Help” button. If the student presses the “help” button without entering any of his solution, the Student Model will receive a “help” packet from the EE. In this case, the Student Model will provide the concept and any information that it has about the student’s knowledge of this concept to the PA. As usual, the PA will decide on feedback/tutoring content and strategy. If the student presses the “help” button after entering a partial solution, this action results in EE evaluating the partial solution. If EE finds errors in student’s work, then “error” packet flow starts between these three components. Otherwise, if the partial solution is successfully evaluated, the SM will notify the PA in which case, the PA responds with a positive feedback followed by an inquiry to the student if he needs help with rest of the solution

There will be instances where the Pedagogical Agent only provides positive feedback that consists of encouraging phrases such as “Good job”, “Congratulations”, etc. The three occasions of such feedback are when the student corrects a problem right after the pedagogical advice, when student’s solution is evaluated successfully without any error or when the student completes the entire homework solution successfully.

PEDAGOGICAL AGENT FLOW CONTROL