Dynamic Group Formation in Mobile Computer Supported

Collaborative Learning Environment

Sofiane Amara

1,2

, Joaquim Macedo

, Fatima Bendella

and Alexandre Santos

Department of Informatics, Faculty of Mathematic and Informatics, University of Sciences and Technology Mohamed,

Boudiaf. B.P 1505 El M’naouer, 31000 Oran, Algeria

Centro Algoritmi, University of Minho, Campus of Gualtar, 4710-057 Braga, Portugal

Keywords: Mobile Learning, Collaborative Learning, MCSCL, Group Formation.

Abstract: Forming suitable learning groups is one of the factors that determine the efficiency of collaborative learning

activities. However, only a few studies were carried out to address this problem in the mobile learning

environments. In this paper, we propose a new approach for an automatic, customized, and dynamic group

formation in Mobile Computer Supported Collaborative Learning (MCSCL) contexts. The proposed

solution is based on the combination of three types of grouping criteria: learner’s personal characteristics,

learner’s behaviours, and context information. The instructors can freely select the type, the number, and the

weight of grouping criteria, together with other settings such as the number, the size, and the type of

learning groups (homogeneous or heterogeneous). Apart from a grouping mechanism, the proposed

approach represents a flexible tool to control each learner, and to manage the learning processes from the

beginning to the end of collaborative learning activities. In order to evaluate the quality of the implemented

group formation algorithm, we compare its Average Intra-cluster Distance (AID) with the one of a random

group formation method. The results show a higher effectiveness of the proposed algorithm in forming

homogenous and heterogeneous groups compared to the random method.

1 INTRODUCTION

The rapid development of wireless communication

and mobile technologies led to the emergence of

Mobile Learning (M-Learning). This new form of

learning allows people to learn anywhere and

anytime thanks to mobility, individuality,

accecibilty, conectivity, and context sensivity of

mobile technologies (e.g. Smartphones, Tablets,

PDA) (Looi et al., 2013). These features allow

providing collaborative, contextualized, customized,

and personalized learning (Baran, 2014).

On the other hand, Collaborative Learning is one

of the important means to improve the

communication skills of learners and to enhance

their knowledge through the exchange of ideas and

experiences. Combining M-Learning with

collaborative learning areas enables the creation of

natural mobile collaboration environments with

face-to-face interactions termed Mobile Computer

Supported Collaborative Learning (MCSCL) (Cortez

et al., 2004). MCSCL allows people to construct

their knowledge collaboratively anywhere, anytime,

and in any context using wireless and mobile

technologies. Thus, many researchers find that

MCSCL represents the next logical step for the

development of collaborative learning field (Boticki

et al., 2010; Caballé et al., 2010).

One of the requirements for an effective

collaborative learning is the appropriate formation of

learning groups. According to (Bekele, 2006),

studies show that the unsuccessful outcomes of

collaborative learning activities are generally due to

failures of the learners grouping. Therefore, the

instructors should pay a great attention to this issue,

in order to provide the necessary conditions for a

succesful collaborative learning.

However, finding the appropriate group for each

learner is a hard and time-consuming task that could

not be well accomplished without computer support

(Hubscher, 2010). In MCSL environments, this task

is more complicated. The grouping process should

not consider only the diversity of learners’ personal

characteristics (age, gender, skills, cultures,

religions, etc), but also the diversity of their learning

behaviours (communication, preferences,

530

Amara S., Macedo J., Bendella F. and Santos A..

Dynamic Group Formation in Mobile Computer Supported Collaborative Learning Environment.

DOI: 10.5220/0005438205300539

In Proceedings of the 7th International Conference on Computer Supported Education (CSEDU-2015), pages 530-539

ISBN: 978-989-758-108-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

movement, etc), and the information related to the

learning contexts. Therefore, our primary objective

is to propose a new approach for forming the best

possible learning groups in MCSCL environments.

In order to achieve this goal, this work identifies

the following research questions:

 RQ1: What is the state of research on this topic?

 RQ2: What are the relevant grouping criteria?

 RQ3: What characteristics of grouping process

contribute for its effectiveness ?

The paper is organised as follows: Section 2

presents the relevant studies with their limitations;

Section 3 describes the proposed approach for group

formation in MCSCL environment with the different

considered grouping criteria; Section 4 presents the

grouping mechanism with emphasis on its

peculiarities; Section 5 describes the system design

and implementation; Section 6 provides an

evaluation of the proposed approach. Finally,

Section 7 presents our conclusions and further work.

2 RELATED WORK

The following list of related works was defined,

evaluated, and analysed using a systematic literature

review (SLR) method (Kitchenham, 2007). We have

presented a more exhaustive description of this SLR

on another paper (Amara et al., 2015). Among more

than 160 found papers that discuss the group

formation problem in MCSCL environments, we

have been able to select only 10 studies that are

considered, by SLR, as the most relevant to this

research problem. These studies are labelled from S1

to S10 and shown in Table 1. The group formation

criteria are classified in three sets: learner’s

characteristics, learner’s behaviours, and context

information.

Study S1 (Huang and Wu, 2011) introduces a

method for collecting some kinds of learning

behaviours and recording them in ubiquitous

portfolios. Then, a systematic grouping mechanism

transforms the collected data into a portfolio grid

and creates a learner similarity matrix. Finally, a

heterogeneous grouping algorithm forms learning

groups using this matrix.

Study S2 (Zurita et al., 2005) presents a MCSCL

environment that supports dynamic changes in the

composition of groups. The authors found that the

dynamic composition of groups contributes with

significant qualitative and quantitative improvement

in learning and social behaviours of learners

(communication, interaction, help, negotiation, etc).

Study S3 (Huang et al., 2010) proposes a

mechanism for analysing the learners’ reading

interests to create learning communities. The study

uses the social platform Del.icio.us to collect the

users’ behavioural data (library’s circulation

records) and recommends partners with similar

interests.

Study S4 (Messeguer et al., 2010) introduces an

approach for group prediction in collaborative

Table 1: Used grouping criteria from the existing approaches.

Id Personal characteristics Learning behaviours Context information

✘

Observing/ Answering quiz/ Interacting/

Moving/ Losing/ Answering questions/

Referencing/ Completing tasks/ Taking

note

Locations of learners and

learning objects

Preferences /Achievement

/ Sociability/ Interests

✘ ✘

✘

Past activities (read books)

✘

Learner’s preferences (place, partners,

preferred subject) / Time spent for learning

Time / Place / Available

neighborhoods

Profile information

✘

Location/ Surrounding objects

S6 Personal background Learner's interaction Location

Learning profile /

Learning styles /

Learning interests

✘

Location

S8 Learner’s interests

Learner’s actions: creating an account/

setting up the profile/ searching for learning

groups/ creating learning groups

✘

Gender / Age /

Motivation / Previous

knowledge

✘ ✘

S10

✘

Helping history

✘

DynamicGroupFormationinMobileComputerSupportedCollaborativeLearningEnvironment

531

learning. The system uses some data of learner’s

behaviours (such as creating, joining, leaving and

destroying groups) to train and test an intelligent

system that could automatically estimate group

membership.

In study S5 (El-Bishouty et al., 2010), the

authors develop a model for finding the best

matched peer helpers for certain tasks. The system

uses RFID tags to detect meaningful surrounding

objects, and create a social knowledge awareness

map for the peer helper based on the detected

objects.

Study S6 (Hsieh et al., 2010) presents a grouping

method based on a mining social interaction. To

evaluate the level of learner’s interaction, wireless

networks are used to measure the distance between

two or more learners in a certain amount of time.

Study S7 (Tan et al., 2010) a dynamic location-

based grouping mechanism is developed. The

proposed process considers the learners’ location

together with other criteria such as learner’s profile,

learning style, and learning interests.

Study S8 (Giemza et al., 2013) presents an

approach for creating informal learning groups. The

used grouping data comes from the campus

environments (personal profiles), location

information, and several types of cloud services (e.g.

Google drive, Brainstormer, and Doodle).

Study S9 (Mujkanovic et al., 2012) demonstrates

the importance of creating groups that exhibit some

desired behaviours. Authors assert that both

individual characteristics and behaviours can be

used to accomplish desired group behaviours.

Study S10 (Yin et al., 2012) presents an

approach that evaluates the level of personal

relationship between learners according to the

frequency of peer helping. The system uses this

evaluation to recommend the appropriate partners

for each learner.

Although the presented studies offer useful

solutions for enhancing the process of learning

group formation, they show some limitations. As

shown in Table 1, the majority of the studies pay a

little attention to the learner’s behaviours. Only S1

proposes an interesting approach for utilizing a set

of behaviours to create learning groups. In addition,

some kinds of learning behaviours are completely

ignored (e.g. communication with instructors and

interaction with learning objects).

Regarding the use of context information as

grouping criteria, one can remark that the location is

the most used criterion. Although the ability of

context awareness offered by the mobile

technologies, the majority of analysed approaches

ignore this type of criteria. Only studies S1, S4, S5,

S6, and S7 proposed solutions to consider some

kinds of context information (time, surrounding

objects, available neighbours).

The dynamic formation of learning groups is

very useful in real world contexts, since the MCSCL

activities are generally exposed to many technical

problems (disconnections, low memory of mobile

devices, etc) and social problems (misunderstanding,

disunion, etc). However, the majority of analysed

studies do not support the dynamic composition of

learning groups.

Another limitation is related to the configuration

settings of the existing grouping processes. The

majority of these processes lack a customized

grouping mechanism. That means that instructors are

unable to select their own grouping choices such as

the nature of groups (homogeneous or

heterogeneous), the number of learners in each

group, and the grouping criteria that could be

considered more appropriate for their learners,

objectives, situations, etc.

3 PROPOSED GROUP

FORMATION APPROACH

In light of the findings obtained from the analysis of

existing approaches, we propose a new approach for

creating suitable learning groups for MCSCL

environments (see Figure 1). The main idea is to

combine the three kinds of criteria (learner’s

personal characteristics, learner’s behaviours and

context information) in a single grouping process

and allow the instructors to customize it according to

different scenarios, activities, learners, needs,

objectives, etc. The following subsections describe

the considered group formation criteria.

3.1 Personal Characteristics

In order to make the group formation process useful

in different learning contexts, the greatest possible

number of grouping criteria should be used. For that,

different personal characteristics are considered

(e.g., age, gender, languages, preferences, skills,

hobbies).

To motivate the learners in their learning, the

proposed system enables them to define the lists of

their preferred partners. And in order to avoid

hindering the learning activities, the system allows

the learners to set the lists of their disliked partners.

The existence of these lists does not necessary mean

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

532

that they will be used. The instructors can or not

consider them.

The proposed approach considers also the

learner’s learning styles. Analysing each learner’s

learning style helps the system to know whether a

learner is: active or reflective, sensing or intuitive,

visual or verbal, sequential or global. To identify

the learning style pattern, each learner has to fill out

a Silverman's index of learning styles questionnaire

(Felder and Silverman, 1988).

Another characteristic considered by the system

is the learner’s health state. The system could

classify the learners into healthy or disabled

learners.

The values of personal characteristics criteria are

generally static, and each learner introduces them to

her/his profile only once. However, some data such

as skills and background knowledge are evaluated

by the system during the learning process.

3.2 Learning Behaviours

Learning behaviours are those actions related to

learning process such as self-motivation, interaction,

communication and satisfaction with the learning

(Dillon et al., 2007).

To ensure a dynamic grouping, the learning

behaviours should be regularly updated. The

following subsections show the list of learning

behaviours used as grouping criteria.

3.2.1 Communication with Partners

In order to evaluate the level of communication

between two learners, two metrics are used: the time

spent in communication and the number of direct

and remote contacts between them. The fact that two

learners are in communication doesn’t mean that

both are active; the system should know which one

of them has initiated the communication.

Based on the evaluation of learner-learner

communication level, the system classifies the

learners into social or introvert.

3.2.2 Communication with Instructor

Similar to the learner-learner communication, the

system should know whether it was the learner who

initiated the communication or the instructor.

The evaluation of learner-instructor

communication level allows the system to classify

the learners based on their autonomy. If a learner

does successfully his tasks with minimum

communication with the instructor, he/she can have

a high level of autonomy.

3.2.3 Interaction with Learning Objects

Learning objects are classified into smart and non-

smart objects. In order to make the non-smart

objects detectable by the mobile devices and allow

the system to control their interaction with the users,

some technologies such as Radio-Frequency

Identification (RFID), or Quick Response (QR)

codes are used to tag them. To measure the level of

interaction between a learner and learning objects,

the system evaluates the number of interactions, and

the time spent in interaction between each learner

and learning object.

3.2.4 Learner’s Movement

The system defines the movement pattern of the

learners by identifying and memorising the different

places they have visited and the learning objects

with which they have interacted. The system could

then classify the learners according to their

movements into moving or passive learners.

3.2.5 Learner’s Preferences

Based on the analysis of learners’ past activities, the

system evaluates and updates continuously the

preferences list of each learner. This list contains the

preferred partners, instructors, activities, learning

objects, places, and times of learning.

3.2.6 Tasks Completion Rate

The instructors ask sometimes their learners to do

some activities (develop projects, study phenomena,

solve problems, etc). In our approach, we assume

that each learning activity is composed of one or

more tasks, and each task could be completed in a

predefined period of time. The learners could be

then classified based on the level of completion of

their past tasks. To enable the system know whether

a learner (or a group) completed a task or not,

learners have to submit their works, pass some tests,

or answer some learning related quizzes. For certain

tasks, the system is unable to evaluate each learner.

In such a case, the evaluation is done manually by

the instructor. At the end of each task, the level of

tasks completion of each learner should be updated.

3.3 Context Information

With the arrival of new wireless and mobile

technologies, the task of collecting and evaluating

the context information becomes possible. However,

the usage of this kind of information to improve the

DynamicGroupFormationinMobileComputerSupportedCollaborativeLearningEnvironment

533

process of group formation is rarely proposed.

Hence, we propose in this approach the use of four

types of context information.

3.3.1 Location of Learners and Surrounding

Learning Objects

MCSCL activities are generally carried out in

informal environments, where learners move freely,

and are not obliged to stay at a given place.

Controlling the learners in such environments is a

hard task or even impossible in some situations.

Nevertheless, the use of mobile technologies enables

tracking the learners and collecting instantaneous

information about them. In our approach, a context

service is used to provide the system with the current

geographical location of the learners and the

surrounding learning objects. The location

information is used also as a way to evaluate some

learners’ behaviours. For instance, to measure the

level of communication between two learners, the

system analyses periodically their geographical

locations. If they stay together in the same place for

a given amount of time, the system considers that

they are in communication and updates their learner-

learner communication level.

3.3.2 Learner and Learning Object

Availability

At a given moment, a learner can be busy, available,

or awaiting for a learning object. Similarly, the

learning objects can be available or in use by

learners. The system must be aware of this context

information (current availability state of the learners

and of the learning objects) in order to avoid

assigning busy learners to new groups, or forming

groups that need to work with already allocated

learning objects.

3.3.3 Learning Progress Level

In order to ensure a dynamic grouping, the

collaborative work of each group should be

periodically controlled. Based on the progress status,

the instructor decides whether a formation of new

groups is or not required.

3.3.4 Time of Learning Activity

Since some learning activities are not similar at

different points in time, this context information

could affect the learning process. Therefore, the

system should be aware about this information

before the formation of learning groups. The time

information is classified into: times of the day

(morning, afternoon, evening, night) and types of

days (weekend, working days, holidays).

4 GROUPING MECHANISM

When a new learner subscribes to the platform,

she/he should use the learner interface to introduce

her/his personal characteristics that are stored in the

learner’s personal profile database (see Figure 1).

In order to continuously collect the learners’

behaviours during the collaborative activities, a data

collection application is installed in the device of

each learner. This application stores frequently the

behaviours data in a set of log files. At the end of

each activity, the system (through the module data

extraction) analyses the log files, extracts the

relevant behavioural information and stores it in the

active database.

Figure 1: Proposed grouping mechanism.

To obtain different context information such as time

of learning and location of learners, the system uses

some mobile technologies such as Global

Positioning System (GPS), Bluetooth, Radio-

Frequency Identification (RFID).

Before starting any collaborative activity, the

instructor, through the module Grouping criteria

and settings, selects the types of criteria (learner’s

personal characteristics, or/and behaviours, or/and

context information). According to the chosen types

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

534

of criteria, the system shows a list of grouping

criteria. The instructor selects then the criteria and

gives a weight for each criterion. In addition, she/he

should define the nature of groups (homogeneous or

heterogeneous), the number of groups, or the

number of learners in each group.

Finally, the grouping algorithm receives the list

of grouping criteria selected by the instructor. It

accesses to the requested databases and interact, if

necessary, with the Context service in order to get

some context information. After obtaining all the

values of necessary grouping criteria, the algorithm

finds the most appropriate learning groups.

Instructor and learners receive then the list of

created groups through their mobile interfaces.

The following subsections describe the main

characteristics of the proposed approach.

4.1 Customized Grouping

The proposed approach gives the instructors a full

freedom to select the type, the number, and the

weight of grouping criteria. In addition, they could

define to the nature of groups (homogenous or

heterogeneous), the number of groups, or the

number of learners per group. Instructors have the

choice between three type of grouping criteria:

learners’ personal characteristics, learners’

behaviours, and context information. They can

choose a single type of them, two, or all together,

depending on the different kinds of learners,

activities, needs, etc. Additionally, the instructors

could specify a weight for each used criteria, which

allows them to give more or less importance to the

various criteria involved in the formation of groups.

This customization in forming groups makes the

grouping mechanism fairly global. It could be used

for any type of learners (young students, secondary

school students, researchers, etc), any type of

activities (developing projects, resolving problems,

etc), and in any learning place (schools, universities,

gardens, museums, campuses, etc).

4.2 Dynamic Grouping

Dynamic grouping means the ability to create

learning groups and change their members at any

moment (Zurita et al., 2005). This ability requires a

continuous update of all the used grouping criteria.

The dynamism of group composition is very

useful in MCSCL environments. Apart from its

ability to change the groups’ members during or

after the end of each learning stage or activity, it

helps the instructors to evaluate the different

learning strategies by using the different grouping

methods in different times and places. In addition,

the dynamic grouping could help a newly arrived

learner to find easily an appropriate learning group.

Furthermore, The majority of MCSCL activities

occur in wide and natural places (gardens, forests,

museums, etc), so, they are generally exposed to a

number of obstacles that could hinder the good

running of the different activities. in certain

situations, these obstacles led to stop the

collaborative activities and destroy the learning

groups. The dynamic grouping represents in those

cases the best solution to quickly restart the

activities with new learning groups.

5 SYSTEM IMPLEMENTATION

For the implementation of the proposed system, we

have used a client–server architecture whose

components are grouped in client, middle and

database tiers (see Figure 2).

5.1 Client Tier

In this tier, users (learners or instructors) use

smartphones or tablets equipped with a web browser

(where the grouping tool should be displayed) and a

set of interface applications (such as GPS, camera,

RFID reader, etc), and some communication tools

(such as Viber and Skype).

The learner’s behaviours (communication level,

visited places, preferred partners, etc) are collected

from specific mobile applications installed on the

user’s device. These applications collect frequently

the data related to the learner’ activities and store

them on a set of log files. At the end of each

collaborative activity, the system through the

module Log analysis analyses these log files to

extract and store the relevant information.

5.2 Middle Tier

This tier represents the application server of the

system. We have chosen Apache Tomcat as a web

server because it is the most flexible, fast and secure.

In this tier two sub tiers are found: presentation and

business tiers. Presentation tier contains Servlets

and Java Server Pages (JSP), which communicate

with each other using the Hypertext Transfer

Protocol (HTTP). The clients (learner or instructor)

view the JSP pages through the mobile browser.

The business tier is composed of a set of

modular components (Java classes) that ensure the

DynamicGroupFormationinMobileComputerSupportedCollaborativeLearningEnvironment

535

good running of the system. The Log analysis

module serves to analyse the log files installed on

the device of each learner as so to extract the

behavioural values and store them in the active

database. The log files are formatted in a way that

facilitates the process of data extraction. Each event

logged in those files mentions the user identifier, the

current location, the current time, and other

information related to the current activity

(communication with partner, interaction with

learning objects, etc).

Figure 2: System architecture.

The Context service module provides the grouping

system with the needed context information (current

time, current locations of learners, location of

learning objects, etc) before starting the grouping

process.

The Data management module manages the data

flow between the client tier and the application

server, and between the application server and the

database tier. This module serves also to normalize

the data to be used by the grouping algorithm.

Additionally, this module allows the users to create

new accounts, to consult and update their data, and

to delete existing accounts.

The Grouping algorithm module receives the

grouping criteria and settings from the instructor’s

browser, obtains the learner’s characteristics and

behaviours from the active database (MySQL

database), and gets the context information from the

context service. The grouping algorithm is

implemented to support both heterogeneous and

homogenous grouping. To form homogeneous

groups, the principle of K-means is used: choosing

K learners as the first members of K groups, and

assigning successively the other learners to closest

groups using the Euclidean distance. To form

heterogeneous groups, the grouping algorithm

searches to maximize the distances within the

learning groups. It creates a similarity matrix

between all the learners, and from this matrix, it

successively searches the farthest pairs of learners to

assign them to the same learning groups. After the

initial formation of a given number of groups, the

algorithm calculates the distance between each

learner and all the created groups in order to assign

him/her to the farthest group.

5.3 Database Tier

This tier serves to store all the data used for learning

processes and the grouping mechanism. The

component of the business tier uses a Java Database

Connectivity Protocol (JDBC) to communicate with

the database. Two kinds of databases are found in

this tier: a main active database (MySQL) installed in

the server side, and temporal databases (log files)

installed on the users’ devices.

6 EVALUATION

6.1 Comparison-based Evaluation

Table 2 shows a comparison between the proposed

group formation approach with the existing studies

presented in Section 2. Through this comparison,

one can remark that the proposed approach is the

only solution that supports the formation of both

types of learning group (homogeneous and

heterogeneous). It is among the few solutions that

consider the three kinds of grouping criteria

(learners’ personal characteristics, learners’ learning

behaviours, and context information). It is one of the

few approaches that enable the user to customize the

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

536

Table 2: Comparison of the proposed group formation approach with the existing approaches.

Study

Groupingtype Groupformationcriteria Groupformationcharacteristics

Heterogeneo

Homogeneou

Personal

characteristic

Learning

behaviours

Context

information

Customized

Dynamic

Continuous

Control of

learners

✔ ✘ ✘ ✔ ✔ ✔ ✘ ✔

✔ ✘ ✔ ✘ ✘ ✔ ✔ ✘

✘ ✔ ✘ ✔ ✘ ✘ ✘ ✘

✘ ✘ ✘ ✔ ✔ ✘ ✘ ✘

✘ ✘ ✔ ✘ ✔ ✘ ✘ ✘

✔ ✘ ✔ ✔ ✔ ✘ ✘ ✘

✘ ✔ ✔ ✘ ✔ ✘ ✔ ✘

✘ ✔ ✔ ✔ ✘ ✘ ✘ ✘

✘ ✘ ✔ ✘ ✘ ✘ ✘ ✘

S10

✘ ✔ ✘ ✔ ✘ ✘ ✘ ✘

Proposed approach

✔ ✔ ✔ ✔ ✔ ✔ ✔ ✔

grouping process, and that provide dynamic

formation of learning groups during the learning

process. Contrarily to the majority of existing

approaches that focus only on the task of group

formation, our approach allows instructors to

permanently control the development of their

learners at several levels (e.g., cognitive, social,

psychological).

6.2 Simulation-based Evaluation

In this subsection, we propose the use of a

simulation method to compare the average intra-

cluster distance (AID) of three grouping methods:

(a) the proposed homogenous grouping approach;

(b) the proposed heterogeneous grouping approach;

the learners of a given group are similar or different

to each other. It provides, therefore, a clear idea

about the level of homogeneity/heterogeneity of

each grouping method. For instance, a low value of

AID implies a great level of homogeneity within the

learning groups, and a high value of AID implies a

great level of heterogeneity. The dataset used in this

simulation was randomly generated from a website

for data generating (http://www.generatedata.com/).

The following group formation criteria were

considered:

 Age (calculated using learner’s date of birth.

Random values from 01-01-1990 to 31-12-2000

were used);

 Gender (male or female);

 Level of communication with learners (random

values from 0 to 6 were used);

 Level of interaction with learning objects

(random values from 0 to 6 were used).

To assess and validate the implemented group

formation algorithm, the simulation process has been

run several times using different settings (e.g.,

different number of learners, different types of

learning group, different number of group formation

criteria).

Figure 3 shows a comparison between the AIDs

of the proposed grouping algorithm (heterogeneous

and homogenous approaches) with the AID of a

random grouping method. The X-Axis represents the

number of learners considered in each phase, and the

Y-Axis represents the AID values. In this first

evaluation process, we have used only one grouping

criteria (the learners age) to evaluate the three group

formation methods. The results show that, whatever

the number of learners (groups) considered in each

simulation session, the resulting AID values of the

proposed heterogeneous grouping approach are

always higher than that of the random method.

Conversely, the AID values of the homogeneous

grouping method are always low compared to the

random method. That implies that the proposed

grouping method forms the most effective groups in

terms of intra-cluster distance.

Figure 4 shows the simulation results of the three

grouping methods considering this time four

DynamicGroupFormationinMobileComputerSupportedCollaborativeLearningEnvironment

537

Figure 3: Average intra-cluster distance of three grouping

methods considering one grouping criterion.

Figure 4: Average intra-cluster distance of three grouping

methods considering multiple grouping criteria.

grouping criteria: age, gender, level of

communication with partners, and level of

interaction with learning objects. Although the

number of criteria increased, the proposed

heterogeneous grouping method forms always

groups with the highest values of AID, while the

lowest AID values are always given by the

homogenous grouping method. That confirms the

effectiveness of the proposed grouping algorithm in

forming the most appropriate groups.

By comparing the AID values resulting from the

both evaluations (using one and multiple criteria), it

is remarked that increasing the number of grouping

criteria results in increasing the AID level, and

therefore, increasing the heterogeneity of learning

groups.

7 CONCLUSIONS

In this paper, a new approach for learning group

formation in Mobile Computer Supported

Collaborative Learning (MCSCL) environments is

presented. First, We have conducted a systematic

literature review (SLR) to analyse the state of

research on this topic. We have found, thanks to this

SLR, that there are no specific grouping criteria that

could be considered ideal. Therefore, we believe that

the choice and selection of such criteria should be

provided by the instructors depending on the

scenarios of learning, the types of activities, the

learning objectives, the needs, the places, the times,

the types of learners, etc. Hence, we have proposed a

customized grouping mechanism that gives the

instructors a full freedom to select the type, the

number, and the weight of grouping criteria. They

could define also the number, the size, and the

nature of groups in terms of

homogeneity/heterogeneity.

The proposed approach considers three types of

grouping criteria: learner’s personal characteristics,

learner’s behaviours, and context information. This

approach does not represent only a grouping tool in

MCSCL environments, but also a very useful mean

for a continuous control of the social, psychological

and cognitive developments of the learners.

The proposed grouping algorithm supports

homogeneous and heterogeneous grouping methods.

To assess how effective this grouping algorithm is,

we have carried out a simulation assessment to

compare the average intra-cluster distance (AID) of

groups created using the implemented algorithm

with the AID of groups created randomly. The

results show high AID values of the groups formed

by the heterogeneous grouping approach, and low

values resulting from the homogenous grouping

approach. That implies a high effectiveness of the

proposed algorithm in creating appropriate groups.

Our future work will deal with developing a

new machine learning approach for criteria

recommendation, to help the instructors for quickly

selecting the proper grouping criteria. Moreover,

evaluating the presented group formation approach

in real world context will help us to extract relevant

information about the relationships between the used

grouping criteria and the corresponding learning and

behavioural outcomes. This extracted information

will be used to develop, train, and test the criteria

recommendation system.

ACKNOWLEDGEMENTS

Sofiane Amara is supported by the program of

Erasmus Mundus UE-MARE NOSTRUM (204195-

CSEDU2015-7thInternationalConferenceonComputerSupportedEducation

538

EM-1-2011-1-ES-ERA MUNDUS-EMA21). This

work has been carried out at Centro Algoritmi,

University of Minho, and partially supported by

FCT -Fundaçãopara a Ciência e Tecnologia- within

the scope of the project PEst-OE/EE/UI0319/2014.

REFERENCES

Amara, S., Macedo, J., Bendella, F., Santos, A., 2015.

Group Formation in Mobile Computer Supported

Collaborative Learning Contexts: A systematic

literature review. Technical report (under

submission). University of Minho.

Baran, E. 2014. A Review of Research on Mobile

Learning in Teacher Education. Journal of

Educational Technology & Society, 17(4).

Bekele, R., 2006. Computer-assisted learner group

formation based on personality traits. Doctoral

dissertation. University of Hamburg, Germany.

Boticki, I., Looi, C.K., Wong, L.H., 2010. Doing

collaboration and learning fractions with mobile

devices. Proceedings of Global Chinese Conference

on Computers in Education 2010 (pp.9-15),

Singapore.

Caballé, S., Xhafa, F., & Barolli, L., 2010. Using mobile

devices to support online collaborative learning.

Mobile Information Systems, 6(1), 27-47.

Cortez, C., Nussbaum, M., Santelices, R., Rodríguez, P.,

Zurita, G., Correa, M., & Cautivo, R., 2004. Teaching

science with mobile computer supported collaborative

learning (MCSCL). Proceedings of the 2nd IEEE

International Workshop on Wireless and Mobile

Technologies in Education (WMTE’04) (pp. 67-74).

Dillon, P., Wang, R., & Tearle, P., 2007. Cultural

disconnection in virtual education. Pedagogy, Culture

& Society, 15(2), 153-174.

El-Bishouty, M. M., Ogata, H., Rahman, S., & Yano, Y.,

2010. Social Knowledge Awareness Map for

Computer Supported Ubiquitous Learning

Environment. Educational Technology & Society,

13(4), 27-37.

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

teaching styles in engineering education. Engineering

education, 78(7), 674-681.

Giemza, A., Manske, S., & Hoppe, H. U., 2013.

Supporting the Formation of Informal Learning

Groups in a Heterogeneous Information Environment.

Proceedings of the 21st International Conference on

Computers in Education. Indonesia: Asia-Pacific

Society for Computers in Education 2013. P.367-375.

Hsieh, J. C., Chen, C. M., & Lin, H. F., 2010. Social

interaction mining based on wireless sensor networks

for promoting cooperative learning performance in

classroom learning environment. The 6th IEEE

international conference on wireless, mobile, and

ubiquitous technologies in education. pp. 219-221.

Huang, J. J., Yang, S. J., Huang, Y. M., & Hsiao, I. Y.,

2010. Social Learning Networks: Build Mobile

Learning Networks Based on Collaborative Services.

Educational Technology & Society, 13(3), 78-92.

Huang, Y., & Wu, T., 2011. A systematic approach for

learner group composition utilizing U-learning

portfolio. Educational Technology & Society. v14 i3.

102-117.

Hubscher, R., 2010. Assigning students to groups using

general and context-specific criteria. Learning

Technologies, IEEE Transactions on,

3(3), 178-189.

Kitchenham, B. A., 2007. Guidelines for performing

systematic literature reviews in software engineering.

version 2.3. Technical Report EBSE-2007-01, Keele

University and University of Durham.

Looi, C. K., Wong, L. H., & Song, Y., 2013. Mobile

computer supported collaborative learning. The

international handbook of collaborative learning, 420-

436.

Messeguer, R., Medina, E., Royo, D., Navarro, L., &

Juárez, J. P., 2010. Group Prediction in Collaborative

Learning. 2010 Sixth international conference on

intelligent environments. pp. 350-355.

Mujkanovic, A., Lowe, D., & Willey, K., 2012. Adaptive

group formation to promote desired behaviours. In

Profession of Engineering Education: Advancing

Teaching, Research and Careers: 23rd Annual

Conference of the Australasian Association for

Engineering Education 2012, The (p. 850). Engineers

Australia.

Tan, Q., Jeng, Y. L., and Huang, Y. M., 2010. A

collaborative mobile virtual campus system based on

location-based dynamic grouping. 2010 10th IEEE

International Conference on Advanced Learning

Technologies. pp. 16-18. IEEE.

Yin, C., Dong, Y., Tabata, Y., and Ogata, H., 2012.

Recommendation of helpers based on personal

connections in mobile learning. 2012 Seventh IEEE

International Conference on Wireless, Mobile and

Ubiquitous Technology in Education. pp.137-141.

IEEE.

Zurita, G., Nussbaum, M., and Salinas, R., 2005. Dynamic

grouping in collaborative learning supported by

wireless handhelds. Educational Technology &

Society, 8(3), 149-161.

DynamicGroupFormationinMobileComputerSupportedCollaborativeLearningEnvironment

539