PROCCESS OF DEFINITION OF LEARNERS STEREOTYPES TO

INTELLIGENT TUTOR SYSTEM BASED ON STRUCTURAL

COMMUNICATION EXERCISES

Robinson Vida Noronha and Clovis Torres Fernandes

Federal Technological University of Parana, Av. Sete de Setembro, 3165, Curitiba Parana, Brazil

Technological Institute of Aeronautics - ITA, Praca Marechal Eduardo Gomes

50 – Vila das Acacias, Sao Jose dosCampos SP, Brazil

Keywords: Learners Stereotypes, Structural Communication, Intelligent Tutor System.

Abstract: The subject of this paper is the forgotten instructional technique called Structural Communication and how

learners stereotypes could be defined to a generic Intelligent Tutor System based on Structural

Communication Exercise. This instructional technique stemmed from teacher’s practice of analysing a

learner’s problem solution to an ill-structured problem. The solution described in this paper is based on

some ideas of convergent and divergent cognitive learner styles. These cognitive learner styles were used to

define a set of twelve basic learner stereotypes. These ideas of learners stereotypes stemmed from the

observation of how learners could employ some domain’s concepts or ideas in a convergence and

divergence way to solve a set of ill-structured problems. This set of learner’s stereotypes is represented by

three independent dimensions or layers and define a Learner Model of a generic ITS based on Structural

Exercise.

1 INTRODUCTION

Hannafin et al. (1999) and Jonassen (2004), among

other researchers, emphasized the lack, necessity

and importance of models or instructional

techniques that could help the Intelligent Tutor

System - ITS developer to represents some

instructional activities based on ill-structured

problems.

To attain this desirable representation, the ITS

developer faces two basic requirements: the

specification of ITS architecture and the selection of

instrucional techniques. This type of challenge

motivates some researches such as those performed

by Arruarte et al, (2003) as well as Heffernan and

Koedinger, (2002) to join instructional techniques to

ITS or Authoring Tools.

In a typical ITS Architecture, the Learner Model

has an important role (Murray, 2003). This Learner

Model could represent several facets and

information about learners. In this context, the ITS

developer could typically define a set of learners’s

stereotypes. For example, Milik et al, (2008) used

two learner’s stereotypes based on spatial ability into

ERM-Tutor. Other example, Parvez and Blank

(2008) defines a set of learner stereotypes based on

Felder-Silverman learning style model (Felder and

Silverman; 1988).

This basic and important ITS requirement was

also researched by Bahar (1999). Bahar identified

some convergence and divergence features in

learners solving ill-structured problems using

Structural Communication – SC (Egan, 1976). SC is

na instructional technique stemmed from the

teacher’s practice of analysing a learner’s problem

solution to a set of ill-structured problems. The

result of this analytical process helps the teacher

select a correct feedback message.

Despite of this, the current state of art of SC

doesn’t report any development of ITS based on this

technique neither how learner stereotypes could be

represented in a computer environment.

This paper describes how Bahar’s convergent

and divergent cognitive learner styles could be used

to define some learner stereotypes to a Learner

Model of a generic ITS based on Structural

Communication exercises.

273

Vida Noronha R. and Torres Fernandes C. (2009).

PROCCESS OF DEFINITION OF LEARNERS STEREOTYPES TO INTELLIGENT TUTOR SYSTEM BASED ON STRUCTURAL COMMUNICATION

EXERCISES.

In Proceedings of the First International Conference on Computer Supported Education, pages 272-277

DOI: 10.5220/0001978502720277

 SciTePress

This paper contains 5 sections. Structural

Communication is summarized in Section 2. Section

3 presents and describes a Learner Model to an ITS

based on Structural Communication execises.

Section 4 analyses the model. Finally, Section 5

presents the conclusion and future works.

2 WHAT IS STRUCTURAL

COMMUNICATION?

Structural Communication is an instructional

technique that individualizes learning, provides

controls for the process by which the learner moves

through the lessons, faces him with challenges to

construct his own multifaceted responses to complex

open-ended problems and ill-structured problems,

analyses these responses and firmly provides

complex, multifaceted, feedback on all relevant

issues revealed by his answer (Egan, 1976).

The Structural Communication technique

involves the development of special units of domain

study. Each learning unit should be structured in

such a way that the learner spends approximately an

hour of study to complete the activities foreseen by

the author. However, the work of the learner is

somewhat analogous to the research of the content

and planning of the structure of an essay or term-

paper type of response - a task that typically takes

many (sometimes many dozens) hours. Thus, the

learner has the opportunity to engage in a much

larger number of creative knowledge-construction

exercises during the time available for study on a

given course. A SC learning unit usually contains

the following sections:

 Intention - This section defines what

should be learned and to what level or

intensity. It supplies a general vision of the

objectives and context for the unit of study.

 Presentation - This section supplies

descriptive information on the subject, possibly

practical exercises or case studies. It can be

composed of text materials, videos,

simulations, computer-based training systems,

hypermedia courses, adaptive hypermedia

systems, electronic games, and site visits,

among other forms.

 Investigation - This section presents a

group of usually 3 or 4 interrelated, challenging

and generally open-ended questions on the

subject of the Presentation. They constitute the

challenge for the learner who responds by

selecting elements from the Response Matrix

presented next.

 Response Matrix - It is a response-

generating instrument formed by a large

number of elements, typically 20, from the

domain under study; they can be sentences that

summarize an idea, key words, concepts or

principles contained in the Presentation. The

learner constructs a response by selecting those

elements that are considered part of a complete

response to the complex question that is being

addressed.

 Discussion - This section is composed of

two parts: a group of " if - then - else " rules

and a series of feedback comments elaborated

by the author, each one associated with one of

the rules. The comments have a constructive

purpose and they discuss in depth the reasoning

used by the learner when selecting or omitting

certain items or subsets of items from the

Response Matrix. They seldom classify a

response as incorrect and never supply a

"correct" response, but rather encourage the

learner to think again and to think deeper and

wider around the issues being addressed.

 Points of View - This last section is used

by a SC exercise’s author to present other

interpretations or conflicting points of view and

to revise some aspects presented earlier. This

section finishes the interaction between the

learner and author, which mimics a virtual

dialogue between them.

One may ask why the potential of researched

methodologies such as Structural Communication

has not been realized by ITS’s developers. One

possible reason for this lack of computer

applications of a theoretically "good idea" is the gap

of Models to represent an SC Unit and SC Domain

Knowledge in a computer environment.

3 LEARNER MODEL TO

STRUCTURAL

COMMUNICATION EXERCISE

The learner stereotypes are defined in this work

using three layers and they are based on the learner’s

last solution to a problem and recorded solution

history. The last learner solution is analysed in the

first layer and clustered based on domain concept

convergence. This solution can be classified in

Convergent (C) or Divergent (D). The Second Layer

analyses the history of solutions and classifies the

learner in Convergent (C), Mixed (M) or Divergent

CSEDU 2009 - International Conference on Computer Supported Education

274

(D). Finally, the Third Layer analyses the history of

solutions and looks for a hidden convergence in all

recorded solutions. The learner is classified as

having presented a Convergent Concept Path (C) or

a Divergent Concept Path (D) if this third layer finds

some convergence in all solutions recorded in the

learner’s solution history.

In this context, an example of a hypothetical

learner “A” is represented in a three dimensional

space in Figure 1. This space can represent any

learner stereotype or learner behaviour. Learner “A”

has a C

onvergent Solution, a Mixed History of

Solutions and a Divergent Concept Path.

Figure 1: Representation of Learner Stereotypes in Three

Dimension Space.

This current paper section is composed of 4

subsections. Section 3.1 describes how Response

Matrix Elements selected by a learner could be

represented in a symbolic expression to be used by a

computer system. This expression records the

presence and absence of some domain concepts in

the Selected Response Matrix Element. Sections

3.2, 3.3 and 3.4 describe a computer algorithm to

analyse this expression. The result of this analisys is

a more precise and refined learner classification.

In this example, a symbolic Response Matrix

with 20 elements is considered. These Response

Matrix Elements are sequential abstract elements

labeled as { F

, F

, ... , F

, F

The same example also considered that Concept

Graphs could be associated with all Response Matrix

elements F

. These Concept Graphs are very similar

to Novak’s Conceptual Maps. (Novak, 1998). Figure

2 illustrates this idea. In this figure, the following

Response Matrix Elements {F

, F

and F

}

were associated with some important

concepts or ideas detached from domain by the SC

exercise author (Noronha, 2005). These concepts or

ideas are labeled as K

nowledge Keyword - KWK in

this context (Noronha, 2005).

Figure 2: Graph Representation of some Response Matrix

Elements.

How can these important concepts or ideas be

represented in a symbolic expression? The next

sections define expression of concepts and show

how learners can be classified in three layer-based

learner stereotype model.

3.1 Expression of Concepts - S

The Expression of Concepts - S

represents the

presence of each KWK in each Response Matrix

element selected by a learner to compose a solution.

The index “1” is used to indicate the presence of

each KWK inside the Matrix Response elements.

For example, the following expressions S

and

represent the elements F

and F

in Figure 2.

The set composed by elements KWK1, KWK3,

KWK6, KWK7 and KWK8 corresponds to element

and another set composed by KWK1, KWK2,

KWK3 and KWK4 corresponds to element F

= 1.KWK1 + 1.KWK3 + 1.KWK6 + 1.KWK7

+ 1.KWK8

= 1.KWK1 + 1.KWK2 + 1.KWK3 + 1.KWK4

All Matrix Response elements can be represented

by a similar expression to those described in this

example.

3.2 First Layer: Classification of

Learners Based on Individual

Analysis of Learners Solution

In this first layer, the problem solution dispatched by

learner must be analysed in an isolated way. Any

solution by the learner could be composed by some

or all Response Matriz elements Fn. For example, a

learner could select the following Response Matrix

elements { F

, F

} to compose his/her solution to

PROCCESS OF DEFINITION OF LEARNERS STEREOTYPES TO INTELLIGENT TUTOR SYSTEM BASED ON

STRUCTURAL COMMUNICATION EXERCISES

275

a challenge or problem defined in SC Intention

Section. These elements are illustrated in Figure 2.

Another learner could select other Response Matrix

elements such as { F

, F

This example is illustrated in the following

Figures: 4 a) e b). The solutions are labeled Sa and

Sb, in these figures. The elements F

and F

compose the solution Sa, whereas the elements F

and F

compose the solution Sb.

Figure 3a) illustrates a solution composed of

Response Matrix elements F

and F

represented

as Venn diagramas. These matrix elements share the

element KWK1. This is represented in Figure 3a) by

means of an overlapping region. In this case, the

solution is classified as Convergent to KWK1.

In contrast, Figure 3b) doesn’t show an

overlapping region. In this sample case, the solution

is classified as Divergent to KWK1, KWK2, KWK3

and KWK5.

Figure 3: Example of Problem’s Solution Analysis.

These analyses could also be conducted using the

Expression of Concepts described in section 3.1.

Each solution expression is created by separately

adding the corresponding index for each KWK. For

example, the solution illustrated in Figure 3a) can be

represented by the following expression of concepts:

4.12.11.2

4.11.1

2.11.1

1118

KWKKWKKWKSSS

KWKKWKS

FFa

++=∧=

The KWK1 has a “2” index because it appears

two times, in F

and in F

. The index analysis of

the concept expression identifies which KWKs have

superior index values. These KWKs indicate a

convergence of ideas or concepts. KWK1 has a

superior index value, in this case. Because of this,

the solution is classified as Convergent because it

converges to KWK1.

The expression of concepts to solution illustrated

in Figure 3b) is represented by :

5.13.12.11.1

1311

KWKKWKKWKKWKS

SSS

FFb

+++=

∧=

The index analysis of expression S

does not

identify index values above “1”. This means that the

solution is not converging on any idea or concept

previously defined by the author. So, this solution is

classified as Divergent to KWK1, KWK2, KWK3

and KWK5.

To summarize, if the concept expression of

solutions had an index value higher than “1”, this

solution is classified as Convergent to KWKs with a

superior index value. If the concept expression did

not have an index higher than “1”, this solution is

classified as Divergent to KWKs with a “1” index.

3.3 Second Layer: Classification of

Learners Based on History of

Problem Solutions

This layer classifies the learner as Convergent,

Divergent and Mixed History of Solutions. If all

solutions recorded in learner’s solutions history were

classed as Convergent, then he/she is clustered as

Learner with Convergent History (C). If all solutions

recorded in learner’s solutions history were classed

as Divergent, then he/she is clustered as Learner

with Divergent History (D). If the learner’s solutions

history has solutions classed both Convergent as

Divergent, then he/she is clustered as Learner with

Mixed History (M).

Although this basic learner classification uses

only three clusters, these clusters can be detailed by

the SC author. For example, a learner clustered as a

Learner with Mixed History could be divided into

smaller categories namely:

• More Convergent, when the amount of

Convergent solutions has higher value than

the amount of Divergent solutions.

• More Divergent, when the amount of

Divergent solutions has higher value than

the amount of Convergent solutions.

• Homogeneity, when the amount of Divergent

solutions is exactely the same as those of

classed as Convergent.

CSEDU 2009 - International Conference on Computer Supported Education

276

3.4 Third Layer: Classification of

Learners Based on Problem

Solution History Path

In this Third Layer, the learner model looks for

some hidden convergence in learner’s solutions

history. For instance, if one learner is classified in

Second Layer as Learner with Mixed History, and he

sequentially presented the following solutions

exemplified in Table 1, is it possible for the Learner

Model to automatically identify some hidden

convergent ideas?

Table 1: Example of Learner with Mixed History record.

S Solution

classification

Expression of Concepts

S0 Convergent to

KWK1

= 2.KWK1 +1.KWK2

+ 1.

KWK4

S1 Divergent S

= 1.KWK1 + 1.KWK3

+ 1.KWK4 + 1.KWK5

S2 Convergent to

KWK2

= 2.KWK2 + 1.KWK4

+ 1.KWK8

S3 Divergent S

= 1.KWK1+1.KWK3

+1.KWK4

The analysis of the set of solutions presented by

learner exemplified in Table 1 indicates one hidden

convergence. KWK4 is present in all learner’s

solutions. In this symbolic example, even though no

solution had been classified as Convergent to

KWK4, this hypothetical learner used the KWK4

element in most or all solutions. This learner is

classed in third layer as a Convergent Concept Path.

To summarize, the analysis of the history of the

learner’s solutions represented in Table 1 gives the

following information:

• The learner’s clustering in the First Layer used

the last solution, S

. This solution is classed as

Divergent.

• The Second Layer clustered this hypothetical

learner as a Learner with Mixed History. This

hypothetical learner oscilated among all types

of solutions. Sometimes he/she presents

Convergent Solutions, sometimes he/she

presents Divergent Solutions.

• Finally, the Third Layer identified a hidden

convergence. This hypothetical learner used the

KWK4 in all solutions presented. This KWK

during the entire solution process may indicate

a possible learner belief.

4 CLASSIFICATION ANALYSIS

The independence of classification i) of individual

solution, ii) from history solution and iii) from

Concept Solution Path allows the specification of

layers that can be used to cluster learners based on a

set of stereotypes including the three layers. These

stereotypes were also defined based on the way the

analytical process can be conducted. The ways are

summarized as follows:

• Individual Solution – The learners can be

classed as Convergent or Divergent. This type

of classification is called “Individual Solution”.

• Individual Solution History – The learners can

be classed as Learner with Convergent History

Solutions, Learner with Divergent History

Solutions or Learner with Mixed History

Solutions. This type of classification is called

“History Solution”.

• Collective Solution History – The learners can

be classed based on Convergent and Divergent

ongoing ideas inside the learner solution

history. This type of classification is called

“Solution Concept Path”.

The combination of these three layers defined in

this paper, allows the identification of 12 basic

learner stereotypes that can be found in an SC

exercise. These stereotypes are presented in Table 2.

Table 2: Basic Stereotypes of Learner Model’s.

Layer of Classification Learner’

s Cluster

First Second Third

Divergent Divergent Divergent

DDD

Divergent Divergent Convergent

DDC

Divergent Convergent Divergent

DCD

Divergent Convergent Convergent

DCC

Divergent Mixed Divergent

DMD

Divergent Mixed Convergent

DMC

Convergent Divergent Divergent

CDD

Convergent Divergent Convergent

CDC

Convergent Convergent Divergent

CCD

Convergent Convergent Convergent

CCC

Convergent Mixed Divergent

CMD

Convergent Mixed Convergent

CMC

Why these stereotypes showed in Table 2 is

important? Because, the learner stereotypes can be

used as guide to formation of feedback message

(Noronha, 2007; Parves and Blank, 2008).

If an ITS running a SC exercise has only one

stereotype then all learners must match with this

stereotype. On the other hand, if there is a high

quantity of stereotypes, then there is more clusters in

this ITS that the each learner could be suitable

PROCCESS OF DEFINITION OF LEARNERS STEREOTYPES TO INTELLIGENT TUTOR SYSTEM BASED ON

STRUCTURAL COMMUNICATION EXERCISES

277

match.

This ability was not defined by creators of SC,

despite it was described by Egan (1976). The

feedback messages could be also adapted based on

learner’s model. For example, a feedback message

could be composed by some “real samples”, “study

of cases”, tables or charts. This message could be

presented to learner clustered as CMC and CCC. A

similar message could be formed by definitions,

explanations and desmonstrations of some domain

concepts. This message could be presented to learner

clustered as DDD or DMD. Clearing some fuzzy

aspects of domain is the purpose of both feedback

messages previously exemplified, but the messages

use distinct ways to accomplish it. The learner’s

stereotypes were used to envelop the feedback

message.

5 CONCLUSIONS

This paper described a model of learner’s

stereotypes definition based on three independent

layers.

These layers were defined based on convergence

and divergence characteristics of learners. These

ideas of learners stereotypes were derived from the

observation of how learners could employ some

domain’s concepts or ideas in a convergence and

divergence way to solve a ill-structured problem.

These main ideas or domain’s concepts are named

KWK and typically they are defined by the author of

SC exercise.

This paper expand some SC characteristics

adding the possibility of employ a Learner Model

during the excecution of SC exercise. In this new

context, feedback message can also be selected and

defined based on learner stereotypes.

The contributions of this paper are the definition

of learner stereotype and a generic learner model

that can be used in Intelligent Tutor Systems based

on SC Exercise. Future work includes some research

questions such as how to define models of feedback

messages based on each of the 12 stereotypes

described in this paper.

REFERENCES

Arruarte, A., Ferrero, B., Fernándes-Castro, I.,

Urretavizcaya, M., Álvares, A. Greer, J., 2003. The

IRIS Authoring Tool. Authoring Tools for Advanced

Technology Learning Environments - Toward cost-

effective adaptive, interactive and intelligent

educational software. Edited by Tom Murray, Stephen

Blessing and Shaaron Ainsworth. Kluwer Academic

Publishers, p. 233-267.

Bahar, M., 1999. Investigations of Biology Students

Cognitive Structure Through Word Association Tests,

Mind Maps and Structural Communication Grids.

Doctoral Thesis. Center of Science Education, Faculty

of Science, University of Glasgow.

Egan, K., 1976. Structural Communication. Fearon

Publishers, Inc.

Felder, R., M., Silverman, L., K., 1988. Learning and

TeachingStyles. Engineering Education, p. 674-681.

Hannafin, M., Lan, S., Oliver K., 1999. Open Learning

Environments, Methods and Models.Instructional-

Design Theories and Models - A new Paradigm of

Instrucional Theory,Lawrence Erlbaum

Associates,Charles M. Reigeluth,vol II, pp. 115-140.

Heffernan, N., T., Koedinger, K. R., 2002. An Inteligent

Tutoring System Incorporating a Model of an

Experienced Human Tutor. Proceedings of 6

International Conference, ITS 2002, France ans San

Sebastian, Spain, June 2002. pp. 596-608.

Jonassen, D., H., 1997. Instructional Design Models fo

Well-Structured and Ill-Structured Problem Solving

Learning Outcomes. Educational Technology:

Research and Development (ETRD), v.45, n. 1, pp.

65-94.

Murray, T., 2003. An Overview of Intelligent Tutoring

System Authoring Tools: Updated Analysis of the

State of the Art. Authoring Tools for Advanced

Technology Learning Environments - Toward cost-

effective adaptive, interactive and intelligent

educational software. Edited by Tom Murray, Stephen

Blessing and Shaaron Ainsworth. Kluwer Academic

Publishers, p. 491-544. ISBN:1-4020-1772-3.

Milik, N., Mitrovic, A., Grimley, M., 2008. Investigating

the Relationship between Spatial Ability and Feedback

Style in ITS. Proceedings of 9

International

Conference, ITS 2008, LNCS 5091, Montreal,

Canada, June 2008, Springer-Verlag, B. Woolf et al

(Eds), p. 281-290.

Noronha, R. V., 2005. Authoring Ideas for Developing

Structural Communication Exercises. YRT Paper.

Proceedings of 12th International Conference on

Artificial Intelligence in Education - AIED 2005. july

18 to 22, 2005. Amsterdam, Netherlands.

Noronha, R., V., 2007. Conceitos e Modelos de Execução

de Exercícios de Comunicação Estrutural. Doctoral

Thesis. Technological Instituto of Aeronautics – SP,

Brazil.

Novak, J. D., 1998. Learning, Creating, and Using

Knowledge: Concept Maps as Facilitative Tools in

Schools and Corporations. Lawrence Erlbaum

Associates, Publishers.

Parvez, S., M., Blank, G., D., 2008. Individualizing

Tutoring with Learning Style Based Feedback.

Proceedings of 9

International Conference, ITS 2008,

LNCS 5091, Montreal, Canada, June 2008, Springer-

Verlag, B. Woolf et al (Eds), p. 291-301.

CSEDU 2009 - International Conference on Computer Supported Education

278