Quality of Group Formation in CSCL Environments

Naseebah Maqtary

1 a

, Abdulqader Mohsen

1 b

and Kamal Bechkoum

Faculty of Computing and Information Technology, University of Science and Technology, Sana’a, Yemen

School of Business and Technology, University of Gloucestershire, Gloucestershire, U.K.

Keywords:

Group, Quality, Collaboration, Framework, Review.

Abstract:

Group Formation (GF) plays a vital role in groupwork performance, for it is the opening phase of the group

development process. Many studies have been conducted to form groups in various scenarios to enhance

collaborative learning. These studies used different clustering techniques, and therefore, the applied evaluation

measures in each study depend on the context of the group formation process. However, there is a lack

of an integrative framework to qualify the overall process of group formation. This paper proposes such a

framework that is composed of layers to tackle each issue related to the GF process. The framework is called

the framework of the Quality of Group Formation (QoGF). QoGF includes three different levels in which every

level has its evaluation measures. These measures are group quality, group formation quality and empirical

quality, which are totaled in an aggregative measure called Total Quality (TQ).

1 INTRODUCTION

Education has improved smoothly through develop-

ing various approaches and technologies (Resta and

Laferri

ere, 2007; Stahl et al., 2006). It has been up-

graded from the individual learning paradigm to col-

laborative learning. With this upgraded paradigm,

learners can gain more knowledge and skills through

learning together from the same learning situation

(Matazi et al., 2014; Resta and Laferri

ere, 2007; Srba

and Bielikova, 2015; Stahl et al., 2006).

Collaborative learning is deﬁned by Rowe et al.

(2010) as an instructional method that is used by a

group of learners to achieve a common goal. The

environment of collaborative learning is either real

or virtual (Dillenbourg, 1999). Collaborative learn-

ing is performed through face to face conversations

and meetings or online using computer tools and

frameworks (Dillenbourg, 1999; Resta and Laferri

ere,

2007; Stahl et al., 2006). An example of such tools

is computer-supported collaborative learning (CSCL)

(Matazi et al., 2014; Rowe et al., 2010; Srba and

Bielikova, 2015; Stahl et al., 2006). CSCL is a ped-

agogical approach that uses networking technologies

to aid the social and instructional interaction among

learners in small groups and learning communities

(Resta and Laferri

ere, 2007; Rowe et al., 2010; Stahl

https://orcid.org/0000-0002-9061-019X

https://orcid.org/0000-0002-3820-918X

et al., 2006).

CSCL has emerged during the mid-1990s. Various

tools have been used and employed to merge col-

laboration within educational activities (Stahl et al.,

2006) such as emails, blogs videoconferencing sys-

tems, content management systems (CMS), and oth-

ers. Focusing on collaborative learning has brought

groupwork to the fore. Many studies in the CSCL

environment have been carried out on administrating

groupwork activities like group formation, monitor-

ing, and evaluation (Sun and Shen, 2013).

Forming a group that collaboratively learns is one

of the most challenging tasks in the CSCLs context.

This topic attracted the interest of several researchers

(Amara et al., 2016; Khandaker et al., 2006; Srba and

Bielikova, 2015).

Group formation (GF) is the opening step of the group

development life cycle. It affects the outcomes of the

group in some way. The process of formation should

be performed with speciﬁc issues in mind. These is-

sues can vary from the group’s special characteristics

to the goal and objectives of the group’s task. In edu-

cational contexts, the learners are grouped according

to their characteristics, the nature of the required task,

and the mechanism of the formation.

However, group formation still has shortcomings in

various perspectives such as attributes, techniques,

and measures, which affect the whole process. GF

process with its speciﬁed attributes and techniques

Maqtary, N., Mohsen, A. and Bechkoum, K.

Quality of Group Formation in CSCL Environments.

DOI: 10.5220/0009391402990306

In Proceedings of the 12th International Conference on Computer Supported Education (CSEDU 2020) - Volume 2, pages 299-306

ISBN: 978-989-758-417-6

299

have been investigated throughly in (Maqtary et al.,

2017), but quality measures are still not covered in

a holistic manner. There is a need to survey the ap-

plied measures in contributed studies. Thus, it will

help to compare these measures and ascertain their

effects on GF. The main motivation behind this study

is to provide a summary of quality measures of group

formation to enhance collaborative learning in CSCL

environments.

This paper proposes a framework for group forma-

tion quality. To achieve this, many steps are traced.

Firstly, surveying the related work and answering the

following questions:

• What are the applied evaluation measures?

• What are the used data, real or synthesized, and

how much are their size?

• Is there a comparison with others’ work?

Secondly, the gaps in the quality of GF in the litera-

ture are speciﬁed based on the given answers. Finally,

an integrative framework is proposed to cover such

gaps.

This paper is organized as follows: the next section

introduces the process of group formation. Related

work is detailed in Section 3. Section 4 is dedicated

to describing ﬁndings while Section 5 shows the pro-

posed framework. Discussing the applicability of this

framework is presented in Section 6. Final remarks

and conclusion is drawn in Section 7.

2 GROUP FORMATION

Group formation is the ﬁrst and opening process of

the group development life cycle in which efforts

should be devoted to ensuring the effectiveness and

efﬁciency of the process (Bonebright, 2010). As men-

tioned above, various research studies were conducted

to explore new provisions in group formation. These

studies attempted to ensure that all group members

are smoothly and efﬁciently achieving the learning

outcomes (Khandaker et al., 2006). To understand

the context of group formation process, some issues

and details should be introduced. The next Subsec-

tion (2.1) highlights the basic concepts of GF such as

the deﬁnition, the attributes, the grouping mode, and

the group size while Subsection (2.2) discusses the

required steps of group formation.

2.1 Preliminaries of GF

Group formation is deﬁned in (Konert, 2014, p.16)

as ”the challenge to optimize learning group for-

mation from a given set of learners, respecting ho-

mogeneously and heterogeneously in simultaneous to

match criteria and aiming for a balanced quality of

the build groups”.

As understood from the deﬁnition mentioned above,

the idea of formation is to organize the learners in

some clusters that need to be balanced. Balanced

groups are groups that achieve their assigned tasks

successfully (Zheng et al., 2018). We add to the deﬁ-

nition the balanced groups should match speciﬁed cri-

teria set by the instructor or the learning situation.

In addressing the high level of collaboration, some

of the current work concentrated on the learner’s at-

tributes/characteristics and their effect on group per-

formance, while other focused on the group’s achieve-

ment and collaboration strategies.

Learner’s attributes represent various aspects of

learner’s readiness to learn. They can be compe-

tences, personality traits, learning style, team role, or

social interaction (Maqtary et al., 2017). Group for-

mation process uses learner’s attributes to decide what

are the most suitable groups that meet the task goal

and requirements.

Another issue related to GF is the grouping mode,

which speciﬁes the type of formed groups. Formed

groups can be homogeneous, heterogeneous, or

mixed depending on the members’ characteristics.

Homogeneous groups have objects that strongly re-

late to each other. Heterogeneous groups have high

diversity between their objects. Mixed groups have

objects that are similar in some attributes and different

in others. As stated in (Graf and Bekele, 2006), ho-

mogeneous groups are better to achieve speciﬁc goals,

while heterogeneous groups are preferred when in-

novative and creative solutions are required. Mixed-

mode needs both homogeneity in some characteris-

tics and heterogeneity in others. The decision about

the best choice of formation mode is based on the task

nature and the required level of collaboration between

learners.

Group size is also another issue that affects group per-

formance but still lacks attention in terms of the num-

ber of signiﬁcant studies that investigate its effective-

ness (Resta and Laferri

ere, 2007). Stahl et al. (2006)

determined that group size is one of the controlling

independent variables that affects collaborative learn-

ing. Kooloos et al. (2011) reported that small group

size is better because it stimulates motivation, cohe-

siveness, development, and cognition. It is worth not-

ing that there is no consensus in literature to deter-

mine the optimal group size in collaborative learning.

However, adequate group size is argued to be ﬁve to

six (Kooloos et al., 2011).

CSEDU 2020 - 12th International Conference on Computer Supported Education

300

2.2 Process of GF

A detailed view of the group formation process is

drawn after surveying existing work. This view may

help researchers to consider the speciﬁc issues of the

process and qualify it. It consists of many steps, as

illustrated in Figure 1. These steps are summarized as

follows:

1. GF Data Gathering: i)gathering the learner and

group attributes, and ii) ordering the priorities of

these attributes according to their effectiveness on

collaborative learning.

2. GF Conﬁguration: specifying the attributes and

mode of the grouping process.

3. GF Execution:it means executing the formation

process using the proposed technique to output the

groups.

4. GF Evaluation: evaluating the output groups us-

ing various metrics.

5. GF Final Approval: giving the instructor the op-

tion of manually ﬁnal groups’ approving.

Figure 1: Steps of GF Process.

This overview of GF gives the most critical as-

pects to be taken into account when considering the

quality of GF. In the next section, the contributions in

this ﬁeld are presented and summarized.

3 RELATED WORK

Based on state of the art, the studies concentrated

on various distance measures used to evaluate each

group. In some cases, the distance measure should

be minimal to clarify the homogeneity within each

group. On the other hand, the bigger the distance

measure is, the better the result of heterogeneous for-

mation is. Distance measures are considered as eval-

uation measures to indicate the acceptance level of

formed groups. However, some studies added further

measures that can be considered as post-performance

measures for the formed groups. They tried to eval-

uate the quality of formed groups through relying on

groups’ performance indicators such as post-tests and

members’ responses to speciﬁc questionnaires.

Furthermore, the experiments of these studies have

been carried out using different datasets. This issue

complicates the comparison between these studies to

ensure the effectiveness of the introduced technique.

This unstandardized evaluation in the process of GF

is the crucial motivation to overview the contributed

factors related to quality and used datasets and con-

clude with sum-up review.

In (Graf and Bekele, 2006), heterogeneous groups

were formed. The heterogeneity was based on the

learner’s score. Average of distance (AD

) in a speci-

ﬁed group was computed through summing the max-

imum and minimum values of members’ attributes.

Then the goodness of heterogeneity (GH) was calcu-

lated by Equation 1.

maxscore(S

) + minscore(S

)

maxscore(S

) − minscore(S

)

1 + Σ



− S

(i)



(1)

where S

j(i)

means the j

learner in i

group.

The higher the GH, the feasible the heterogeneity

in the group is. Graf and Bekele (2006) used ﬁve

datasets of 100 learners and checked the quality of

the grouping through the GH only. It also measured

its scalability by executing the algorithm on a dataset

of 512 learners. No comparison was held with others.

Ounnas et al. (2007) proposed a general framework

for evaluating the quality of the GF process. The re-

searchers proposed a metrics framework to evaluate

constraint satisfaction-based group formation. The

framework based on ontologies. They used formation

quality (FQ) and some constraints violations (NCV)

parameters, which were built on average and standard

deviation (SD) to evaluate their framework. The pro-

posed FQ dealt with goals satisfaction. As researchers

stated that from the learning viewpoint, the group for-

mation quality is a multi-dimensional concept which

implies the formation efﬁciency besides the groups’

performance indicators. As shown in Figure 2, the

framework is composed of the following three parts:

1. Formation metrics

• Constraint Satisfaction (Group, Cohort).

• Perceived formation satisfaction (Individual,

Group, Cohort).

2. Productivity metrics

3. Goal satisfaction metrics

• Goal satisfaction quality (Group, Cohort).

• Formation quality (Group, Cohort).

The framework was not empirically implemented and

no comparison was held with others.

Quality of Group Formation in CSCL Environments

301

Figure 2: Group Formation Quality Ounnas et al. (2007).

Ho et al. (2009) used particle swarm optimization

(PSO) technique to form heterogeneous groups. This

study used one real dataset of 61 learners. However,

no comparison was held with others. The required

formation quality was obtained by maximizing the

objective function, as shown in Equation 2.

max

= w

a=1

b=a+1

(−

CPT

−CPT

)

+ w

j=1

(DIF

) + w

j=1

(INT

/Σ

x=1

i j

(2)

where CPT is the total competences of the j

group,

INT is the overall interaction among students in the

group, and DIF is the summation of the style dif-

ferences among students in the j

group.

Yannibelli and Amandi (2011) used evolutionary al-

gorithm to form heterogeneous groups with balancing

roles as illustrated in Equations 3, 4 and 5.

nr(G

,r) =











1 r is naturally played by one member of G

−2 r is not naturally played in G

−p r is naturally played by p members in G

(3)

nb(G

) = Σ

r=1

nr(G

,r)

(4)

max

∀G∈C



b(G) =

i=1

nb(G

)



(5)

where nr(G

,r) calculates the balance levels of the

roles in each group, as shown in Equation 3 and

nb(G

) computes the average of balances in all groups

as presented in Equation 4. The last Equation 5 calcu-

lates the best population that generates the balanced

groups (nb should equal to nine roles based on Bel-

bin’s model). The study conducted many experiments

using ten synthesized datasets. It also compared its

performance by the execution time with the exhaus-

tive method (EM).

Moreno et al. (2012) formed intra-heterogeneous and

inter-homogeneous groups using genetic algorithm

(GA). The study used a real, local dataset composed

of 135 learners. It compared the GA with the other

two algorithms; random method (RM) and (EM). It

used D

as an evaluation measure, as shown in Equa-

tion 6. The less the distance D

is, the more the inter-

homogeneous group.

= Σ

g=1

− X

g,1

)

+ (C

− X

g,2

)

+ ... + (C

− X

g,M

)

, (6)

where C

is the mean of each attribute in the group i,

is the mean of each attribute in all individuals.

In the same vein, Tien et al. (2013) improved GA to

form heterogeneous groups. The study used the ﬁt-

ness function that was represented in Equation 7.

= ω· GP + (1 − ω)·(1 − GP

), (7)

where GP, GP

and ω respectively denote the mean

value of all groups, the standard deviation of all

groups, and the weight of GP, GP

. The mean of the

relative closeness to the optimal solution is indicated

using GP

, as shown in Equation 8.

q=1

(

s=1

s,q

−V

−)

·ω

q=1

(

s=1

s,q

−V

·ω

q=1

(

s=1

s,q

−V

−)

·ω

(8)

As stated in the study, the proposed algorithm gave

slightly better solutions compared to other algorithms,

which were simple GA (SGA) and (RM). It used ﬁve

simulated datasets with various volumes to evaluate

the proposed GA.

Amara et al. (2016) used the Average Intra-cluster

Distance (AID) without mentioning the equation. The

bigger the measure is, the more heterogeneity the

cluster is. They used simulated dataset without spec-

ifying its volume, and no comparison was held with

others

Acharya and Sinha (2018) formed mixed groups.

They used the measure E, which is called the sum

of square errors to ensure the homogeneity, as shown

in Equation 9.

E = Σ

i=1

Σa ∈ C

|a − m

, (9)

where a is an individual, and m

is the mean of the

group C

. Also, heterogeneity was expressed by Equa-

tion 10.

H(G

) =

, (10)

where x is the number of intervals that represent a cer-

tain attribute, and r is the number of group’s mem-

bers. Comparison was held with other studies includ-

ing (Graf and Bekele, 2006) and (Christodoulopoulos

and Papanikolaou, 2007). The study used one real

dataset with 72 learners.

4 FINDINGS

As summarization showed in Table 1, group forma-

tion quality in the literature lacks a comprehensive

CSEDU 2020 - 12th International Conference on Computer Supported Education

302

Table 1: Summary of GF Quality in Literature Work.

Author and

publication

year

Evaluation Equations Dataset Comparison

with

Others

Homo

Hete

Mixed

Local

Real

Synth.

Graf and Bekele

(2006)

- Equ. 1 - 5 X - -

Ounnas et al.

(2007)

- 1 X X -

Ho et al. (2009) - Equ. 2 - 1 X - -

Yannibelli and

Amandi (2011)

- Equ. 3 & 4 & 5 - 10 - X EM

Moreno et al.

(2012)

- Equ. 6 - 1 X - EM & RM

Tien et al. (2013) - Equ. 7 & 8 - 5 - X SGA & RM

Amara et al.

(2016)

* *

- 1 - X -

Acharya and

Sinha (2018)

- - Equ. 9 & 10 1 X - *

Not available.

Not applied.

quality framework that considers all expected con-

texts. Each study used its proposed measure and,

in the best cases, ran some comparisons with oth-

ers to evaluate their proposed metrics and techniques.

The only introduced quality framework was detailed

in (Ounnas et al., 2007) and presented in Figure 2.

However, it is still a proposal framework and not im-

plemented or evaluated yet. Moreover, it is not suit-

able for all group formation models since it was for-

mulated based on ontologies. Its measures and levels

concentrate on empirical performance.

According to the above discussion, inadequate con-

tributions are apparent in the ﬁeld of quality metrics

that measure the quality of the group formation pro-

cess from different viewpoints.

Therefore, it is essential to implement a comprehen-

sive quality framework for the group formation pro-

cess. The framework should give directive indica-

tors about the groups’ formation process for the in-

structors. There is no need to wait until the end of

the groups’ works to evaluate their formation. In-

stead, some measures need to be proposed to assess

the formed groups before beginning the actual work.

The concentration should be on how well the groups

were formed rather than how well they will perform.

These measures will indicate the success of the for-

mation process. In this context, the following ques-

tions may arise: How can the group formation process

be evaluated? what are the most important factors that

indicate the success of the formation process?

5 QUALITY OF GROUP

FORMATION FRAMEWORK

(QoGF)

Group formation in CSCL is essential, especially in

large cohorts and short-term groups. To qualify the

formation process, there is a need for quality met-

rics. Here, a comprehensive framework quality of

group formation (QoGF) is proposed. Based on the

surveyed literature, there is a lack of comprehensive

quality metrics that deal with all speciﬁcations of the

formation process. QoGF is composed of three levels

to reﬂect various aspects of group formation. Inten-

tionally, the order of the levels reveals that the quality

should be considered while forming groups. Group

quality, formation process quality, and empirical qual-

ity are the levels of QoGF. The following subsections

deﬁne these levels and present their applied evalua-

tion measures.

5.1 Levels of QoGF

As Figure 3 shows, QoGF has three nested levels. The

most inner level is the group quality (GQ), while the

most outer one is the empirical quality (EQ). Forma-

tion process quality (FPQ) is in the middle of the pro-

posed framework. Below sections deﬁne each level

and discuss its role in the GF process.

5.1.1 Group Quality (GQ)

GQ is the inner level, which ensures two perspectives:

each group’s quality and satisfaction of assigned con-

straints. The ﬁrst perspective is the quality of each

group, which is called intra-class quality. Various

distance measures can be used to examine the intra-

class quality, such as Davis-Bouldin Index (DBI),

Dunn, etc. GQ measures evaluate the homogeneity

and compactness inside each group according to the

clusters’ validity indicators. Some of them also mea-

sure the separability between groups. Choosing the

measure is based on the nature of the data and the con-

text of the formation process. The second perspective

is the constraints satisfaction, which veriﬁes if the

formation process fulﬁlls the predeﬁned constraints

such as the task’s nature, the instructor’s constraints,

and the learner’s preferences. These constraints are

selectively applied depending on the context of the

GF process. They cover the various required conﬁgu-

rations as discussed below.

1. Task’s nature deals with different aspects that rely

on the context of the required task. These aspects

are various, including grouping mode (Homoge-

neous, heterogeneous and/or mixed), size (vari-

able or ﬁxed) and/or priorities of attributes.

2. The instructor’s constraints give opportunities for

the instructor to set up his constraints, such as ex-

cluding somebody from a speciﬁc group or solv-

ing the problem of unassigned learners (orphans)

to any group at the end of the process.

3. The learner’s preference takes in to account the

preferences of learners as some tasks take care of

Quality of Group Formation in CSCL Environments

303

learners’ satisfaction (e.g., preserving the friend-

ship between the learners).

5.1.2 Formation Process Quality (FPQ)

The purpose of FPQ is evaluating the goodness and

balance of the whole process of formation. It is di-

vided into two parts: inter-class quality and algorithm

quality.

1. Inter-class quality evaluates whether the whole

formed groups are balanced. It means that groups

are similar in satisfying the formation constraints

and are in the same level of quality. To measure

such issue, any balance factor can be applied.

2. Algorithm quality evaluates the time and space

complexity of the used algorithm.

5.1.3 Empirical Quality (EQ)

EQ is the outer level of the QoGF framework. It

evaluates the success of the groups’ formation (e.g.,

the groups’ performance) according to the achieved

goals. Abnar et al. (2012) reported that most of the

GF studies did not evaluate the real outcomes after

groups’ task completion. The only thing researchers

measured is comparing the value of quality factors

they deﬁned for the groups formed by their algorithm.

This assumption leads us to consider this type of qual-

ity. It is measured through reporting and analyzing the

responses of all groups members to the predesignated

questionnaires that investigate many factors such as

satisfaction, performance, goal achievement, and ad-

vancement in collaborative learning such as discussed

in (Abnar et al., 2012).

Figure 3: Proposed Quality Framework for GF (QoGF).

5.2 Evaluation Measures

As presented in the QoGF framework, many evalu-

ation measures are available to be used among the

different levels of the framework. These levels are

applicable to any chosen grouping mode, which are:

homogeneous, heterogeneous, and mixed. The fol-

lowing list clariﬁes each mode with its possible eval-

uation measure.

1. Homogeneous mode in a speciﬁc attribute can

be achieved through minimizing the difference

between the values of that attribute in the same

group. Forming homogeneous groups means

that similarity/homogeneity should be high intra-

group and dissimilarity/heterogeneity should be

also high inter-groups. Many existed validation

measures are used to qualify this type of groups

such as homogeneity and separation scores, sil-

houette width, redundant score and WADP (Chen

et al., 2002).

2. Heterogeneous mode can be evaluated through

maximizing the number of various values of a spe-

ciﬁc attribute that should be heterogeneous in the

same group. Heterogeneous formed groups are

those groups with high dissimilar/heterogeneous

intra-group and high similar/homogeneous inter-

groups. One of the most attractive measures

that are used to evaluate these groups is K-

complementarity measure. K-complementarity

measure veriﬁes the fulﬁllment of all roles in the

each group. It is borrowed from forming groups

of tutors in the educational context, as presented

in (Laﬁﬁ et al., 2014).

3. Mixed mode means that formed groups are com-

posed of objects with some attributes that are

homogeneous, and other attributes are heteroge-

neous. In this mode, evaluation measures are

combined from measures of both modes; homo-

geneous and heterogeneous.

Below subsections introduce the measures GQM,

FPQM, and EQM of the GQ, FPQ, and EQ levels,

respectively, with consideration of all modes of group

formation: homogeneous, heterogeneous and mixed.

Besides, another measure called total quality measure

(T QM) is introduced to ﬁnalize the GF quality pro-

cess. T QM is calculated based on the outcomes of

the GQ, FPQ, and FPQ measures.

5.2.1 Group Quality Measure

GQM evaluates the fulﬁllment of GQ criteria; intra-

class and constraints satisfaction. It also varies based

on the chosen grouping mode. In homogeneous

mode, any intra-class quality measure can be used.

Also, the percentage of satisﬁed constraints selected

by the instructor and learners should not be violated.

These two criteria are formulated by groups

intra−class

and constraints, respectively. They should be max-

imized in homogeneous mode to fulﬁll the quality

CSEDU 2020 - 12th International Conference on Computer Supported Education

304

of the GQ level. On the other hand, the heteroge-

neous mode can use the same criteria of homogene-

ity with the inverse of groups

intra−class

. This inverse

is required to reﬂect the difference inside the formed

groups. Finally, in the mixed-mode, the same mea-

sures of both homogeneity GQM

homo

and heterogene-

ity GQM

hete

are applied as a preferred setting that is

speciﬁed by the instructor. The group quality level is

totally measured by GQM, as shown in the Equation

11. The more the value of GQM is, the better the re-

sults are obtained.

GQM =











constraints + groups

intra−class

Homogeneous groups

constraints +

groups

intra−class

Heterogeneous groups

GQM

homo

+ GQM

hete

Mixed groups

(11)

5.2.2 Formation Process Quality Measure

In FPQ, the QoGF middle level, different criteria are

chosen to evaluate the inter-class quality and algo-

rithm complexity. Inter-class quality should measure

the balance and equilibrium among all formed groups

by applying any assigned balance measure, which

may be affected by the formation context and instruc-

tor preference. It is expressed by proc

inter−class

, and it

should be maximized. Algorithm complexity can be

expressed by algo

complexity

. The algorithm complex-

ity should be minimized to gain better performance.

The FPQM value should be maximized to achieve

better results and calculated using Equation 12.

FPQM =

algo

complexity

+ proc

inter−class

(12)

5.2.3 Empirical Quality Measure

The outer level of QoGF is to model the empirical per-

formance of formed groups. As discussed previously,

it is measured by the learners’ improved performance

and questionnaires about their satisfaction of achieve-

ment. All these indicators should be quantiﬁed using

suitable measures (chosen − measures) and should be

maximized to reﬂect better results. EQM is expressed

by Equation 13

EQM = max(chosen − measures)

(13)

5.2.4 Total Quality Measure

The total quality measure T QM expresses the quality

of the whole GF process as shown in Equation 14. It

is the summation of the qualities of the QoGF levels

with preferred weights set by the instructor. The value

of this measure should be maximized to record the

best results.

T QM = GQ ∗ w

+ FPQ ∗ w

FPQ

+ EQ ∗ w

(14)

6 DISCUSSION

As presented in Section 5, this paper introduces an in-

tegrative framework, QoGF, for the process of group

formation in different cases and constraints. In com-

parison with the literature, previously conducted stud-

ies applied the GF process on specially chosen con-

texts, while QoGF gives an umbrella for applying

the process by enabling the instructors to set the pro-

cess conﬁguration according to the required collabo-

rative context. The GF process conﬁguration includes

the chosen grouping mode, used intra-group measure,

balance among formed groups, algorithm feasibility,

outcomes of the formed groups, and other assigned

preferences and constraints. Further, QoGF offers the

capability of evaluating the achievement of such con-

ﬁgurations through different levels and assigned mea-

sures.

The contribution of QoGF is its ability to ﬁnalize the

GF process using a quantitative measure that indicates

the overall process quality. In addition, it can be con-

sidered as a basis that allows comparison among dif-

ferent scenarios and settings to form groups using var-

ious grouping techniques.

7 CONCLUSIONS AND FUTURE

WORK

The group formation process is an important step that

affects the other steps of the group development life

cycle in CSCL environments. Several studies were

conducted to automatically form groups in different

contexts. They used various quality measures to eval-

uate the process. This paper reviews the existing

studies and summarizes their contributions according

to quality metrics and used datasets. After analy-

sis of these studies, shortcomings in GF quality were

recorded. Therefore, an integrative quality framework

was proposed to alleviate the lack of quality standards

in GF.

This framework includes three layers to reﬂect the na-

ture of the process. It is composed of group qual-

ity (GQ), formation process quality (FPQ), and em-

pirical quality (EQ). Besides that, different evaluation

measures are formulated to permit the applicability of

qualifying the GF process through various levels and

contexts.

For future work, this framework needs to be imple-

mented and evaluated to ensure its suitability to most

contexts of formation in CSCL environments. Ac-

cording to the applied datasets, it is preferable to have

a benchmark dataset that can be used in the context of

group formation.

Quality of Group Formation in CSCL Environments

305

REFERENCES

Abnar, S., Orooji, F., and Taghiyareh, F. (2012). An evolu-

tionary algorithm for forming mixed groups of learners

in web based collaborative learning environments. In

Technology Enhanced Education (ICTEE), 2012 IEEE

International Conference on, pages 1–6. IEEE.

Acharya, A. and Sinha, D. (2018). A ”mixed” strategy for

collaborative group formation and its learning outcomes.

Journal of Educational Technology Systems, 46(4):440–

462.

Amara, S., Macedo, J., Bendella, F., and Santos, A. (2016).

Group formation in mobile computer supported collab-

orative learning contexts: A systematic literature re-

view. Journal of Educational Technology & Society,

19(2):258–273.

Bonebright, D. A. (2010). 40 years of storming: a histor-

ical review of tuckman’s model of small group devel-

opment. Human Resource Development International,

13(1):111–120.

Chen, G., Jaradat, S. A., Banerjee, N., Tanaka, T. S., Ko,

M. S., and Zhang, M. Q. (2002). Evaluation and com-

parison of clustering algorithms in analyzing es cell gene

expression data. Statistica Sinica, pages 241–262.

Christodoulopoulos, C. E. and Papanikolaou, K. A. (2007).

A group formation tool in an e-learning context. In 19th

IEEE International Conference on Tools with Artiﬁcial

Intelligence (ICTAI 2007), volume 2, pages 117–123.

IEEE.

Dillenbourg, P. (1999). What do you mean by collaborative

learning. Collaborative-learning: Cognitive and compu-

tational approaches, 1:1–15.

Graf, S. and Bekele, R. (2006). Forming heterogeneous

groups for intelligent collaborative learning systems with

ant colony optimization. In International Conference on

Intelligent Tutoring Systems, pages 217–226. Springer.

Ho, T.-F., Shyu, S. J., Wang, F.-H., and Li, C. T.-

J. (2009). Composing high-heterogeneous and high-

interaction groups in collaborative learning with parti-

cle swarm optimization. In Computer Science and In-

formation Engineering, 2009 WRI World Congress on,

volume 4, pages 607–611. IEEE.

Khandaker, N., Soh, L.-K., and Jiang, H. (2006). Student

learning and team formation in a structured cscl environ-

ment. FRONTIERS IN ARTIFICIAL INTELLIGENCE

AND APPLICATIONS, 151:185.

Konert, J. (2014). Interactive Multimedia Learning: Using

Social Media for Peer Education in Single-Player Edu-

cational Games. Springer.

Kooloos, J. G., Klaassen, T., Vereijken, M., Van Kuppeveld,

S., Bolhuis, S., and Vorstenbosch, M. (2011). Collabo-

rative group work: Effects of group size and assignment

structure on learning gain, student satisfaction and per-

ceived participation. Medical Teacher, 33(12):983–988.

Laﬁﬁ, Y., Bendjebar, S., and Zedadra, A. (2014). A k-

complementarity technique for forming groups of tutors

in intelligent learning environments. Journal of comput-

ing and information technology, 22(2):115–130.

Maqtary, N., Mohsen, A., and Bechkoum, K. (2017). Group

formation techniques in computer-supported collabora-

tive learning: A systematic literature review. Technology,

Knowledge and Learning, pages 1–22.

Matazi, I., Messoussi, R., and Bennane, A. (2014). The de-

sign of an intelligent multi-agent system for supporting

collaborative learning. In Intelligent Systems: Theories

and Applications (SITA-14), 2014 9th International Con-

ference on, pages 1–8. IEEE.

Moreno, J., Ovalle, D. A., and Vicari, R. M. (2012). A

genetic algorithm approach for group formation in col-

laborative learning considering multiple student charac-

teristics. Computers & Education, 58(1):560–569.

Ounnas, A., Millard, D. E., and Davis, H. C. (2007). A

metrics framework for evaluating group formation. In

Proceedings of the 2007 international ACM conference

on Supporting group work, pages 221–224. ACM.

Resta, P. and Laferri

ere, T. (2007). Technology in support of

collaborative learning. Educational Psychology Review,

19(1):65–83.

Rowe, J. P., Shores, L. R., Mott, B. W., and Lester,

J. C. (2010). Integrating learning and engagement in

narrative-centered learning environments. In Interna-

tional Conference on Intelligent Tutoring Systems, pages

166–177. Springer.

Srba, I. and Bielikova, M. (2015). Dynamic group for-

mation as an approach to collaborative learning support.

IEEE Transactions on Learning Technologies, 8(2):173–

186.

Stahl, G., Koschmann, T. D., and Suthers, D. D. (2006).

Computer-supported collaborative learning.

Sun, G. and Shen, J. (2013). Teamwork as a service:

a cloud-based system for enhancing teamwork perfor-

mance in mobile learning. In Advanced Learning Tech-

nologies (ICALT), 2013 IEEE 13th International Confer-

ence on, pages 376–378. IEEE.

Tien, H.-W., Lin, Y.-S., Chang, Y.-C., and Chu, C.-P.

(2013). A genetic algorithm-based multiple character-

istics grouping strategy for collaborative learning. In In-

ternational Conference on Web-Based Learning, pages

11–22. Springer.

Yannibelli, V. and Amandi, A. (2011). Forming well-

balanced collaborative learning teams according to the

roles of their members: An evolutionary approach. In

Computational Intelligence and Informatics (CINTI),

2011 IEEE 12th International Symposium on, pages

265–270. IEEE.

Zheng, Y., Li, C., Liu, S., and Lu, W. (2018). An improved

genetic approach for composing optimal collaborative

learning groups. Knowledge-Based Systems, 139:214–

225.

CSEDU 2020 - 12th International Conference on Computer Supported Education

306