• No results found

2.2 Phase 2: Design a Tutorial Conversation

2.2.1 Steps 2.1 and 2.2

For the Oscar ACITS prototype, it was decided to reuse the tutorial developed for the Oscar PCITS prototype, which delivers an SQL revision tutorial.

Table 10.2.Domain-independent Pseudo-code Adaptation Algorithm Applied to the FS model

Let:

the question scores be qSNS, qINT,qACT, qREF, qSEQ, qGLO, qNEU,

the student learning style classes be cSI, cAR cSG

the student scores be sSI, sAR, sSG,

the student question scores be aSI, aAR, aSG (where SI, AR and SG are the classes of learning style

dimensions),

the set of winning student question scores be W.

IF ((cSI==NEU) AND (cAR==NEU) AND (cSG==NEU))

THEN

{adapt_to_class(NEU);} // if all neutral learning styles, choose neutral adaptation

ELSE // calculate student question scores

{

IF (cSI==SNS) THEN aSI=qSNS*sSI

ELSE IF (cSI==INT) THEN aSI=qINT*sSI

ELSE aSI=0;

IF (cAR==ACT) THEN aAR=qACT*sAR

ELSE IF (cAR==REF) THEN aAR=qREF*sAR

ELSE aAR=0;

IF (cSG==SEQ) THEN aSG=qSEQ*sSG

ELSE IF (cSG==GLO) THEN aSG=qGLO*sSG

ELSE aSG=0;

W = get_max_list(aSI, aAR, aSG); // get list of winners – maximum scores

IF size(W) == 1 THEN // if only one winner adapt to that class

{

IF MAX(aSI, aAR, aSG) == aSI THEN adapt_to_class(cSI);

ELSE IF MAX(aSI, aAR, aSG) == aAR THEN adapt_to_class(cAR);

ELSE adapt_to_class (cSG);

}

ELSE // if >1 winner adapt to max student score in W

{

X = get_max_list_from(sSI, sAR, sSG,W);

IF size(X) == 1 THEN // if only one winner adapt to that class

{

IF MAX(sSI, sAR, sSG) == sSI THEN adapt_to_class(cSI);

ELSE IF MAX(sSI, sAR, sSG) == sAR THEN adapt_to_class(cAR);

ELSE adapt_to_class (cSG);

}

ELSE // if >1winner adapt to max question score in W

X = get_max_list_from(qSI, qAR, qSG,W);

IF size(X) == 1 THEN // if only one winner adapt to that class

{

IF MAX(qSI, qAR, qSG) == qSI THEN adapt_to_class(cSI);

ELSE IF MAX(qSI, qAR, qSG) == qAR THEN adapt_to_class(cAR);

ELSE adapt_to_class (cSG);

}

ELSE // if more than one winner adapt to Random in W

adapt_to_class(RANDOM(W))

} }

Key:

SI=Sensor/Intuitive; AR=Active/Reflective; SG=Sequential/Global

NEU=Neutral; SNS=Sensory; INT=Intuitive; ACT=Active; REF=Reflective; SEQ=Sequential; GLO=Global

For step 2.1, the original capture of the tutorial conversation is described in Chapter 7, Section 4.1. The tutorial scenario consisted of ten tutorial questions (Table 7.2) with resources that support the adaptation of tutoring to different learning styles and a multiple choice question (MCQ) test.

For step 2.2, the captured tutorial conversation was structured by applying the 3- level model of a tutorial conversation, as described in Chapter 7, Section 4.2. Then the tutorial questions were mapped to the generic question templates, as described in Chapter 7 Section 4.3.

The resulting tutorial conversation was documented in the tutorial conversation blueprint document. Appendix 3 shows an excerpt of the tutorial conversation blueprint.

2.2.2 Step 2.3: Map Tutorial Questions onto the Generic Teaching Material Categories

In Step 2.3, the captured tutorial conversation was tailored to make it suitable for an adaptive tutorial. This was done by mapping the tutorial questions to the generic teaching material categories (Chapter 9, Table 9.3). Next, this mapping was

reorganised to highlight the available adaptations per question, and learning styles requiring additional adaptive teaching material. Further interviews with human tutors were undertaken to gather more material to improve the adaptations in the SQL tutorial, such as introductions and more examples. The additional resources were then documented in the tutorial conversation blueprint document, agreed with human tutors and the mappings updated. Table 10.3 shows the final mapping of the SQL revision tutorial to learning style adaptations, derived from the mapping to teaching material categories. Every question also has a Neutral learning style adaptation which includes a mixture of styles.

As many tutor material categories as possible were included in the design of each tutorial question. It is recognised that to make a coherent learning experience, some tutorial questions may not lend themselves to adaptation (e.g. Q3 in Table 10.3 has only two adaptations other than Neutral). This demonstrates the strength of the adaptation algorithm in considering both the individual question adaptations and student learning styles strengths over algorithms that require adaptive material for all learning styles, or those that adapt to just one dimension (see Chapter 9, section 3).

Table 10.3. Learning Style Adaptations in the SQL Revision Tutorial Question Neutral Active Reflective Sequential Global Sensor Intuitive

Introduction     Q1       Q2        Q3   Q4        Q5        Q6        Q7       Q8       Q9       Q10      

2.2.3 Step 2.4: Score Tutorial Questions for Adaptation to Each Learning Style

In Step 2.4, each tutorial question was assigned a score for every learning style representing the number (or strength) of opportunities for adaptation to that learning style. This was done for each question by simply counting the number of times each category of teaching material related to a learning style was available. Where no adaptations existed for a learning style, the score assigned was zero. The neutral learning style was always assigned a score of zero. Table 10.4 shows the final scores for the SQL revision tutorial.

Table 10.4. Question Adaptation Scores

Question Active Reflective Sequential Global Sensor Intuitive

Introduction 0 0 1 2 1 1 Q1 4 1 1 1 3 3 Q2 2 1 1 1 2 2 Q3 2 0 0 0 2 0 Q4 5 3 3 1 5 4 Q5 9 5 11 11 9 11 Q6 3 2 2 1 3 3 Q7 4 2 2 0 4 3 Q8 3 2 2 0 3 2 Q9 7 3 7 7 7 3 Q10 4 2 2 0 4 2

2.2.4 Step 2.5: Script Conversational Agent Natural Language Dialogue for each Tutorial Question using the 3-Level Model

The development of scripts for the InfoChat CA (Convagent Ltd., 2005) is fully described in Chapter 7, Section 4.4. For the Oscar ACITS prototype, whilst the

organisation of the scripts over three levels into separate contexts stayed the same, multiple versions of the CA scripts representing different adaptations were required. During this step, for each learning style the Oscar PCITS prototype scripts were amended to match the mapped teaching material categories for that learning style (as documented in the tutorial conversation blueprint). For example for the Active learning style, CA scripts for tutorial questions based on the hints question template (see Chapter 7, Table 7.2) were amended to replace any detailed explanations of theory with hints based on practical examples and exercises.

For each tutorial question, a CA script name was assigned to each learning style and to the Neutral learning style. Where a question score was zero for a learning style (i.e. no adaptation existed) the neutral CA script was assigned. The script names and scores were recorded for inclusion in the tutorial knowledge base component, which will be described in Section 2.3.

2.3 Phase 3: Construct the ACITS Architecture

The Oscar ACITS architecture proposed in Chapter 9, shown in Figure 10.1, was adopted. Oscar ACITS‟ modular structure allowed several of the components

developed for the prototype Oscar Predictive CITS (described in Chapter 7, Section 5) to be reused, speeding up development.

Controller Graphical User Interface Conversational Agent Learning Styles Adapter Scripts User Student Model Tutorial Knowledge Base

Figure 10.1. Oscar ACITS Architecture

Controller

The controller manages the tutorial and communicates with all components. The controller developed for the Oscar PCITS prototype (Chapter 7, section 5) was

reused but changed. For the Oscar ACITS prototype, additional functionality relating to dynamically selecting adaptations was required, as follows:

At the start of each tutoring question, retrieve the student learning style scores and tutorial question scores, and pass to the learning styles adapter

component.

On receiving the start context from the learning styles adapter, pass to the CA and start the conversation.

Apart from managing individual adaptations (as described above), the

controller‟s management of the tutorial remained the same as in the Oscar PCITS prototype.

Graphical User Interface (GUI)

For the Oscar ACITS prototype, the GUI component created for the Oscar PCITS prototype (Chapter 7, section 5) was reused and required no changes.

Student Model

For the Oscar ACITS prototype, the student model developed for the Oscar PCITS prototype (described in Chapter 7, Section 5) was extended by adding the Session table to record the adaptations. This enables detailed information about the adaptive student learning experience to be analysed. The Oscar ACITS prototype student model class diagram is shown in Figure 10.2.

Student PK RegNo Password Act Ref Seq Glo Sns Intu Vis Vrb LastQuestion ILS PK RegNo Q1..Q44 Act Ref Seq Glo Sns Intu Vis Vrb Results PK RegNo Q1..Q12 Score Q21..Q32 Score2 Session PK RegNo PK ModID PK Qno Adaptation 0..1 1..1 1..1 0..1 0..* 1..1

Figure 10.2.Student Model Class Diagram

The Session table has a record for every tutorial question a student has attempted, and records the adaptation applied.

Conversational Agent (CA)

The CA module from the Oscar PCITS prototype, described in Chapter 7, Section 5, was reused for the Oscar ACITS prototype and required no changes. The design of the CA scripts followed step 2.5 of phase 2 of the Oscar ACITS

methodology, described in Section 2.2.4. Tutorial Knowledge Base

This module was developed by following phase 2 of the Oscar ACITS methodology, as described in Section 2.2. For the Oscar ACITS prototype, the tutorial knowledge base developed for the Oscar PCITS prototype (described in Chapter 7, Section 5) was extended to include a new table recording the adaptation details for each tutorial question, as shown in Figure 10.3. The adaptation scores are used by the Learning Styles Adapter module to determine the best adaptation for a tutorial question. The adaptation start contexts are passed to the CA at the start of each tutorial question.

Modules PK ModID Qcount TestQcount Tests PK ModID PK Qno Qtext Atext Btext Ctext Dtext Answer Adaptations PK ModID PK Qno PK LS Score StartContext 1..1 0..* 1..1 1..*

Figure 10.3.Tutorial Knowledge Base Class Diagram

The modules table was amended to remove the StartContext column, which is now recorded in the adaptations table.

The adaptations table has a record for each learning style for each module tutorial question, recording the adaptation score and the name of the CA script to start with (as there may be several linked contexts for a question, see section 2.2.4).

The tutorial knowledge base is related to (but not linked to) the CA scripts that deliver the tutorial conversation.

Learning Styles Adapter

The learning styles adapter module contains the implementation of the adaptation algorithm. The learning styles adapter receives the question adaptation scores for each learning style from the tutorial knowledge base and the student learning style

scores from the student model (via the controller) and determines the best adaptation for each tutorial question. This module was developed by following phase 1 of the Oscar ACITS methodology.

This completes the development of the Oscar ACITS prototype. The next section will show some example dialogue that demonstrates how the tutoring is adapted for different learners.