Optimizing the Usability of User Interfaces based on Multi-objective

Evolutionary Algorithms

Marwa Hentati

1 a

, Abdelwaheb Trabelsi

2 b

and Adel Mahfoudhi

3 c

National School of Engineering, Sfax University, Sfax, Tunisia

College of Computing and Informatics, Electronic University, Dammam, Saudi Arabia

College of Computers and Information Technology, Taif University, Taif, Saudi Arabia

Keywords:

User Interface, Optimization, Usability, Evolutionary Algorithm, Model-Driven Engineering.

Abstract:

Solving the software system problems using optimization algorithms stands for an intrinsic area of research

whose aim is to ﬁnd an optimal solution according to a set of conﬂicting objectives. One of the most prominent

problems is optimizing the software quality such as usability of user interfaces following the model-driven en-

gineering (MDE). One of the main challenges of MDE process is identifying the highly-usable model accord-

ing to a set of desired usability aspects. Although models may be equivalent from the functional viewpoint,

they may differ from the non-functional perspectives. Besides, they do not fulﬁl the same usability properties.

In this context, we adressed this issue by combining the power of model engine and the optimization algo-

rithms. In this study, we propose to integrate a multi-objective evolutionary algorithm at the conceptual level

of the MDE process. It allows to ﬁnd an optimal (or near-optimal) model from a large search space according

to a set of usability aspects and taking into account the context of use.

1 INTRODUCTION

Usability of the software systems is no longer a lux-

ury, but rather an intrinsic factor of the success or fail-

ure of software systems (Seffah et al., 2006). Seeking

to promote a highly usable user interfaces has trig-

gered the appearance of some research studies taking

into account the usability aspects of software systems

(Abrah

ao and Insfran, 2017). In fact, a large num-

ber of these research works consider usability aspects

only after the full implementation of systems. In this

case, all changes introduced to the system design are

costly in terms of resources and time. The basic ob-

jective of other research works resides in enhaning us-

ability in the intermediate artefacts. Since the trace-

ability between these artefacts and the implimented

system is not well deﬁned, taking into account the us-

ability aspects at the conceptual level may not ensure

the usability of the ﬁnal system (Abrah

ao and Ins-

fran, 2017). Supporting the model-driven engineer-

ing (MDE) process may alleviate these problems ow-

ing to its intrinsic traceability mechanisms (da Silva,

https://orcid.org/0000-0003-3832-8602

https://orcid.org/0000-0002-1416-6867

https://orcid.org/0000-0001-9522-8099

2015). Among the most outstanding resarch works

in this context, we mention the Cameleon Reference

Framework (CRF)(Calvary et al., 2003). It organizes

the user interface (UI) development process into four

levels of abstraction: At the highest level, the task

& concept model (T&C) gathers the logical activi-

ties (tasks). Then, the abstract user interface (AUI)

model groupes a set of presentation-units which are

modality independent. After that, at the concrete user

interface (CUI) model gathers a set of computing-

platform dependent interactors. Finally, the ﬁnal user

interface (FUI) model expresses the UI in terms of

implementation technology. From this prespective,

we support the idea that improving usability at the

conceptual level leads to improve the ﬁnal user in-

terface (Hentati et al., 2016a). Recently, several re-

search works have centred around supporting some

usability aspects from the early stages of MDE pro-

cess (I

niguez-Jarr

ın et al., 2020). Nevertheless, there

are still certain defﬁciences quoting in particular the

lack of details about generating the target model with

an optimal usability (Hentati et al., 2016b). Hence,

the search-based techniques stand for an appropiate

solution to settle the software engineering (SE) prob-

lems by exploring the search space. Multi-objective

optimization algorithms, in particular the evolution-

Hentati, M., Trabelsi, A. and Mahfoudhi, A.

Optimizing the Usability of User Interfaces based on Multi-objective Evolutionary Algorithms.

DOI: 10.5220/0010581902770284

In Proceedings of the 16th International Conference on Software Technologies (ICSOFT 2021), pages 277-284

ISBN: 978-989-758-523-4

277

ary algorithms, are applied widely to the majority of

the SE problems including requirements management

(Pitangueira et al., 2015), testing (Afzal et al., 2009)

and maintenance (Bavota et al., 2014) disregarding

the usability aspects (Harman and Jones, 2001).

The remaining of this paper is organised as fol-

lows. Section 2 displays an overview of related

works. Section 3 exhibits the proposed process to op-

timize the usability at the conceptual level. Section 4

introduces a case study illustrating the feasibility and

the obtained results. Finally, Section 5 wraps up the

conclusion and provides new prespectives for future

research.

2 RELATED WORKS

The central focus of this section is to summarize the

state of the art works and the efforts performed in

the ﬁeld of software engineering via MDE approach.

We are basically intersted in some proposals that are

considered relevant owing to their weight in related

works.

Authors in (Mkaouer et al., 2020) proposed an ap-

proach for solving the SE refactoring problems. This

problem is formulated as a multi-objective problem

using the NSGA-III algorithms. The basic target re-

sides in ﬁnding the optimal refactoring solution refer-

ring to a set of quality attributes. In the same context,

authors in (Ouni et al., 2013) suggested to reformu-

late this problem as a single-objective optimization

using an evolutionary algorithm to search the opti-

mal sequence of refactoring operations. The opera-

tions quality is improved by minimizing the number

of code smells.

Recently, several studies have addressed the

Search-Based Model Transformation (SBMT) sub-

ﬁeld which combines the search techniques and the

model transformation engines (Boussa

ıd et al., 2017).

We state, as an example, the MOMoT approach

(Marrying Optimization and model Transformations)

(Fleck et al., 2017). The basic objective is to trans-

form a component model from a class diagram us-

ing optimization techniques by means of a Multi-

Objective Evolutionary Algorithm (MOEA). To op-

timize the quality of the obtained model, authors at-

tempted to minimize coupling and to maximize cohe-

sion as objective functions.

Since the small screens devices are unceasingly

used, an approach is set forward in (Raneburger et al.,

2015) aiming to consider the generation of the graph-

ical user interface from the Discourse-based commu-

nication model as an optimization problem. This ap-

proach rests on the branch-and-bound algorithm for

transforming the communication model into the struc-

tural model by selecting the optimal rule transforma-

tion. To accomplish this goal, three objective func-

tions are formulated corresponding to the maximiza-

tion of the use of available space, minimization of

the number of navigation steps and minimization of

scrolling the user interface.

The analysis of the above mentioned studies indi-

cates that there are still such drawbacks to be over-

come as considering the functional properties. Most

of the proposed approaches search the optimal solu-

tion disregarding the usability optimization problem.

In order to respond to this need, we opted for the opti-

mization of the usability of user interfaces at the con-

ceptual level following the MDE process by means of

a multi-objective evolutionary algorithm.

3 USABILITY OPTIMIZATION

PROCESS USING AN

EVOLUTIONARY ALGORITHM

The basic goal of this research is to generate a con-

crete user interface (CUI) model with an optimal us-

ability from an AUI model. We adressed the trans-

formation of the AUI to the CUI as a usability opti-

mization problem. As the usability displays a contra-

dictory effect, we opted for multi-objective evolution-

ary algorithm. In prior research studies (Hentati et al.,

2018) (Hentati et al., 2019), we tackled the generation

of CUIs considering all alternatives related to abstract

components which determine the search space.

Even through evolutionary algorithms (EAs) are

generally based on the same generic framework, sev-

eral requiremnts need to be taken into account to set-

tle the usability optimization problem. The elaborated

lagorithm begins with the generation of an initial CUI

that is added to the current population (Algorithm 1,

line 10). For each iteration, all CUIs in the current

population are assessed depending on the proposed

objective functions (Algorithm 1, line 14).

One of the basic mechanisms of the evolutionary

algorihm is to generate new solutions by means of the

reproduction operators: the mutation operator (Algo-

rithm 3) and the crossover operator (Algorithm 4).

After i iterations, the search ends and returns a set

of optimal solutions which stand for the approximate

pareto front (PF

approx

In order to select the optimal solution from

approx

, we propose to use the Knee point strat-

egy (Bechikh et al., 2011) depicted in Algorithm 2.

The Knee point solution is a vector composed of the

best objective values among the solutions produced

ICSOFT 2021 - 16th International Conference on Software Technologies

278

Algorithm 1: The evolutionary algorithm for

the usability optimization problem.

1 Input

2 Population size : α

3 Mutation rate: β

4 Crossover rate: σ

5 Elitism rate: γ

6 Iterations number: δ

7 Output PF

Approx

Begin Generate the initial

CUI ;

8 POP ←initial CUI;

9 Iteration number i ←1 ;

10 while i < δ do

11 /* Objective functions*/ Evaluate

the usability of each CUI in POP (using

objective functions 1, 3 and 6);

12 /* Select best solutions*/

Select (α ∗ γ) best CUI in POP ;

13 Add selected CUI in population POP’;

14 /* Eliminate unfeasable CUI */

foreach CUI in POP’ do

15 if constraints are not satisﬁed

(equations 9 and 11) then

16 Delete CUI from POP’;

17 end

19 end

20 /* Mutation */ foreach (α ∗ β) CUI

21 Apply mutation operator (Algorithm

3);

22 Add CUI mut in POP’;

24 end

25 /* Crossover */

26 Select randomly (α ∗ σ) CUI from POP’

for crossover;

27 foreach (α ∗ σ) CUI do

28 Select two CUIs as parents;

29 Apply crossover operator (Algorithm

4);

30 Add crossovered CUIs in POP’ ;

32 end

33 Increment the iteration number i ← i + 1

34 end

in each iteration. To ﬁnd the optimal trade-off, these

solutions are assessed using the trade-off worthiness

metric introduced by (Rachmawati and Srinivasan,

2009). The closest one to the obtained ’Knee point

solution’ is then chosen as the ﬁnal optimal solution.

Algorithm 2: ’Knee point’ strategy.

1 Input Aproximatif Pareto front PF

Approx

(Output Algorithme 1)

2 Output Optimal CUI

3 Begin

4 foreach Objective O

5 Select best values V

relative to O

determine the knee point solution

CUI

;

6 end

7 foreach CUI

in PF

approx

8 Calculate the distance between CUI

and

CUI

(equation 12) ;

9 end

10 Select the optimal CUI;

11 End

3.1 Objective Functions

The usability optimization problem includes search-

ing for the most usable CUI among a set of candi-

dates, which corresponds to a large search space. In

this context, we propose to search the CUI with an

optimal usability according to three objective func-

tions. Each of them stands for a usability sub-

characteristic which can be either maximized or

minimized. These usability sub-characteristics are

adopted from an agreed usability model proposed by

(Ammar et al., 2016). Indeed, this model is extended

from the ISI/IEC 25000 standard to evaluate the us-

ability at the early stage of the devlopement process

(ISO, 2005). The suggested objective functions are

portrayed below:

3.1.1 Maximizing the Learnability

Optimizing the usability necessitates maximizing the

value of the learnability sub-characteristic as identi-

ﬁed in equation 1.

= Maximize(PR) (1)

In fact, the learnability is speciﬁed by the ’Prompting’

attribute. Equation 2 depicts the PR metric associated

with the prompting attribute.

PR =

∑

i=1

SuppIn f o(x

)

(2)

where PR is the proportion of labels that exhibits sup-

plementary information, x

is a label in a UI, n is the

number of text ﬁelds that needs a speciﬁc structure

and SuppIn f o = 1 if label presents supplementary in-

formation, 0 otherwise.

Optimizing the Usability of User Interfaces based on Multi-objective Evolutionary Algorithms

279

3.1.2 Minimazing the Understandability

As for the Understandability sub-characteristic, it is

characterized by a set of attributes having a relation-

ship with cognitive load

. Hence, minimizing the

value of understandability yields the optimization of

the CUIs usability. In fact, we suggest minimizing

two attributes, namely ’Breavity’ and ’Information

density’, equation 3.

= Minimize(BR,ID) (3)

BR and ID metrics are determined respectivelly in

equations 4 and 5.

BR = MinDistance(x,y) (4)

where BR is the number of steps needed to move from

x to y, x, y ∈ (inter f ace).

ID =

∑

i=1

Element (5)

where n denotes the total number of elements per UI.

3.1.3 Maximizing the Operability

The usability of concrete models is optimized through

maximizing the Operability sub-characteristic

grounded on two attributes, which are ’error preven-

tion’ and ’user control action’, equation 6.

= Maximize(ERP,UCA) (6)

EPR and UCA metrics are speciﬁed respectivelly in

equations 7 and 8 .

ERP =

∑

i=1

list(x

)

(7)

where ERP corresponds to proportion of primitive

lists used for each input element with a set of enu-

merated values, x

stands for an input element, n rep-

resents the set of input elements with enumerated val-

ues, list = 1 if the input element with enumerated val-

ues uses primitive list, list = 0 otherwise.

UCA =

∑

i=1

(8)

where UCA expresses the proportion of user control

actions including cancellability and undoability ac-

tions, x

indicates a UI action element and n denotes

the total number of elements per UI.

The cognitive load (or workload) is related to the human

memory properties. (Chanquoy et al., 2007). According to

(Bastien and Scapin, 1993), the higher the cognitive load

is, the higher is the risk of errors, lateness, exhaustion, and

dissatisfaction of the user during interaction with the sys-

tem.

3.2 Optimization Constraints

The majority of optimization problems, restrictions

are imposed referring to the problem speciﬁcation.

Threfore, the solution acceptability is guaranteed only

after the satisfaction of the optimization constraints.

Moreover, it is difﬁcult to generate a user interface

with optimal usability for all user’s proﬁles and run

on all plateforms (Gajos et al., 2010).

In fact, taking into account the context of use is

a basic purpose in the MDE process. We have taken

into account the context of use deﬁned by (Calvary

et al., 2003) who introduced the context as the triplet

<user, platform, environment>. Since the expres-

siveness of environment is conﬁned to the concep-

tual level, we opted for the platform (considering the

screen size) and the user (considering the user expe-

rience level). Feasible solutions are chosen for the

next population only if they satisfy the proposed con-

straints:

∑

(ε + a

) ≤ availableSpace (9)

with:

availableSpace = X

resolution

∗Y

resolution

(10)

where a

refers to the area of a component i of the

CUI, ε indicates the minimum space that separates

two components, and availableSpace represents the

screen area of the device to be involved.

A user with limited experience (or novice user)

needs to be well guided, informed and protected from

errors during interaction with the user interface. In

fact, usability attributes like prompting (PR), and er-

ror prevention (ERP) are related to the user, especially

to his level of experience. As a matter of fact, we

set forward an optimization constraint based upon the

level of the user’s experience. The target lies in ﬁl-

tering solutions where the avredge (AVG) values of

PR and ERP are under a threshold (V

). The later

is ﬁxed by the ’conceptor’ considering the experience

level of the target population. The equation 11 dis-

plays the proposed optimization constraint relative to

the user level experience:

AV G(PR,ERP) ≥ V

(11)

3.3 Reproduction Operators

Reproduction operators invade the research space

through the progress of candidate solutions. The

derivation of new CUIs is accomplished by the mu-

tation and crossover operators. The production of de-

scendants plays an crucial role in the convergence of

evolutionary algorithms to a global optimum.

ICSOFT 2021 - 16th International Conference on Software Technologies

280

3.3.1 Mutation Operator

Instead of applying standard mutation operators that

might not be thoroughly convenient for our problem,

we foreground a personalized mutation operator de-

pected in Algorithm 3. This operator offers a kind of

diversity within the population and permits the opti-

mal solution to escape from local optimum.

We propose to mutate one or a set of elements of a

current CUI by (1) changing a component to another

alternative and/or (2) adding a component to the CUI.

Changing components in the CUI is fulﬁlled by re-

placing them with alternative concrete elements. It is

signiﬁcant to point out that the number of elements to

be mutated rests on the mutation probability P

ﬁxed

before launching the mutation operator.

As for the addition of elements, we propose the

addition of control elements, like ”validate” or ”can-

cel” in the CUI, after verifying that it is impossible to

cancel or to validate information in it.

Algorithm 3: Mutation operator.

1 Input CUI for mutation

2 Mutation probability P

3 Output mutated CUI

4 Begin

5 if no ’valid’ element in CUI then

6 Add valid element in CUI;

7 end

8 if no ’cancel’ element in CUI then

9 Add cancel element in CUI;

10 end

11 S

CUI

← size of CUI;

12 N

is the number of elements to be mutated;

13 N

← S

CUI

∗ P

;

14 foreach (N

) element in CUI do

15 Randomly select a concrete component

Comp from CUI;

16 Mutate Comp with an alternative;

17 Generate a new CUI mut ;

18 Add CUI mut in the population;

19 end

20 End

3.3.2 Crossover Operator

The crossover operator is applied at the level of pan-

els of two randomly chosen CUIs. It can guarantee

that all elements, which are gathered in a panel, are

migrated. Moreover, the ”one-cut crossover” is cho-

sen because the ”multiple-cut crossover” can disturb

optimal solutions and the algorithm can converge to

sub-optimal solutions. The crossover operator is de-

pected in Algorithm 4.

Algorithm 4: Cossover operator.

1 Input Two CUIs as parents: CUI P1 et

CUI P2

2 Output Two CUIS as childs: CUI Ch1 et

CUI Ch2

3 Begin

4 Cut CUI P1 in two parts CUI P11 and

CUI P12 ;

5 Cut CUI P2 in two parts CUI P21 and

CUI P22 ;

6 Combine CUI P11 with CUI P22 as

CUI Ch1;

7 Combine CUI P21 with CUI P12 as

CUI Ch2;

8 Add CUI Ch1 and CUI Ch2 in the

population;

9 End

4 AN ILLUSTRATIVE CASE

STUDY

The main aim of this case study is to emphasize the

feasibility of the proposed usability optimization pro-

cess. This case stands for a BTTS (Buying Train

Ticket System). This system runs on terminals of pas-

sengers (smart phone, tablet, laptop, etc.). The UI of

this system needs to create a feeling of comfort and

ease of use to upgrade the user satisfaction degree re-

gardless of the computing device. Numerous tasks

are involved in BTTS like How to travel, Buy tickets,

Discount and beneﬁts, Train times etc. We are basi-

cally concerned with the generation of the CUI of Buy

tickets task.

4.1 Generation of the Initial CUI

To better clarify the workability of the proposed us-

ability optimization process, the case study started

with the transformation from an abstract model to a

concrete model. This ordinary transformation did not

consider the usability aspects in a speciﬁc context of

use. This system is implemented in mobile device-

platformes with small screens size, and interacted by

novice users having a limited experience of using a

buying systems in general and BTTS in particular.

The ordinary transformation takes as input the AUI

model and produces the initial CUI model. Yet, the

generated CUI fulﬁlled their functional objectives dis-

regarding the usability aspects.

Optimizing the Usability of User Interfaces based on Multi-objective Evolutionary Algorithms

281

Table 1: The obtained objective values.

Iter min max min max min max

1 0.375 0.375 0.452 0.452 0.476 0.476

10 0.389 0.672 0.452 0.452 0.277 0.476

20 0.389 0.618 0.447 0.452 0.334 0.561

30 0.477 0.584 0.423 0.561 0.399 0.561

40 0.277 0.573 0.387 0.611 0.408 0.533

50 0.290 0.583 0.415 0.611 0.412 0.601

60 0.402 0.573 0.415 0.611 0.427 0.624

70 0.453 0.588 0.384 0.629 0.508 0.682

80 0.521 0.592 0.384 0.638 0.513 0.747

90 0.577 0.603 0.362 0.638 0.513 0.853

100 0.601 0.622 0.374 0.638 0.513 0.871

110 0.601 0.725 0.362 0.647 0.622 0.851

120 0.639 0.752 0.362 0.647 0.648 0.851

126 0.647 0.753 0.362 0.647 0.652 0.851

4.2 Obtained Results of the Optimal

Transformation

The ﬁndings of running the usability optimization

process at AUI to CUI transformation are outlined

in Table 1 which summarizes the objective functions

values after each step of 10 iterations. We infer

that the values of these objective functions are sta-

ble from the 126th iteration. This last iteration groups

8 CUIs which represent the approximate Pareto front

(PF

approx

). The values of each of these CUIs are il-

lustrated in Table 2. The ”min” and ”max” values,

highlignted in Table 1, denote respectively the lowest

and highest values of the objective functions, for each

CUI produced in a population. Departing from these

values, the ”Knee point” solution (CUI

) is identi-

ﬁed resting on the best CUI for each objective func-

tion. Indeed, the ”Knee point” solution is deﬁned as:

< f

=0.753; f

=0.362; f

=0.871 >.

In order to specify the ﬁnal optimal CUI among

the 8 CUIs of the PF

approx

, we applied the Euclidean

distance permetting to identify the closest CUIs to the

”Knee point” solution as identiﬁed in equation 12.

d(CUI

,CUI

) =

∑

(CUI

−CUI

)

(12)

After computing the distance of each CUI with the

”Knee point” solution, we realized that CUI5 is the

ﬁnal optimal CUI, which has the following values: <

=0.705; f

=0.374; f

=0.851>.

Figure 1 exhibits the ﬁnal optimal CUI generated

by the proposed usability optimization process for

mobile devices. In this context, small screens are

not able to present all the information and the user

needs to scroll in order to see the interface. In fact,

the high information density has a negative effect on

Table 2: The CUIs values of PF

approx

and their distances

from ”Knee point” solution.

approx

Distance

”Knee

point”

IUC1 0.753 0.366 0.652 0.203

IUC2 0.753 0.647 0.804 0.292

IUC3 0.661 0.362 0.683 0.201

IUC4 0.753 0.381 0.655 0.254

IUC5 0.705 0.374 0.851 0.043

IUC6 0.659 0.362 0.804 0.128

IUC7 0.647 0.644 0.871 0.306

IUC8 0.661 0.647 0.804 0.303

the user’s performance and eventually on the usability

of the CUI. To settle this problem, the CUI elements

are distributed into three windows (Figure 1 a, b and

c). Consequently, the association of windows instead

of panels may ensure the optimization of the informa-

tion density attribute (ID).

Yet, the window association negatively inﬂuences

the brevity attribute (BR) in view of the increas-

ing number of steps required to accomplish the task.

Since the CUI is distributed into three windows, two

steps are required to carry out the task. Therefore,

as the number of steps is less than 3 (an acceptable

upper bound), we infer that the brevity attribute is op-

timized.

The PR attribute is materialized in the optimal

CUI by adding supplementary information about

the data entry format. Indeed, this information

guides users as well as facilitates their interaction

with the system, where a speciﬁc format is ex-

pected to be entered such as the format of the dates

(DD/MM/YYYY).

At another level, users of the BTTS have to be

protected against errors during data entry. In fact,

to accomplish this target, input elements with a list

of options are transformed into lists, such as the

drop-down list (UIDropDownList), the radio button

(UIRadioButton) the checkbox (UICheckBox) and

the combo box (UIComboBox). The presence of

these elements allows to have a CUI with an optimal

error prevention (ERP).

Additionally, the optimization of the user con-

trol actions (UCA) attribute is materialized by adding

control buttons such as the validation and cancella-

tion button. Indeed, the ”Valid” button enables the

validation of data entered by the user. The ”Cancel”

control button enables the user to interrupt an action

or a treatment in progress.

ICSOFT 2021 - 16th International Conference on Software Technologies

282

Figure 1: The optimal CUI in the mobile context.

4.3 Learned Lessons

The case study allows the additional assessment of

the feasibility of optimizing the usability of user in-

terface from conceptual models into MDE process.

Furthermore, the suggested approach permits to take

into acount usability at an early stage of the devel-

opment process. This may help avoid the usability

problems once the system is operated. In addition,

we inferred that the number of iterations as well as

the performance of this process are affected not only

by the optimization parameters but also by the qual-

ity of the initial CUI as well as the optimization con-

straints linked to the context of use. The found results

can be regarded as promising. Indeed, some improve-

ments can be introduced with respect to adding some

usability attributes as well as validating them with an

empirical study.

5 CONCLUSION

The paper elabrates a usability optimization process

into MDE paradigme. This process yields three in-

trinsic steps: the ﬁrst one concerns the generation of

the intial CUI without taking into account the usabil-

ity aspects. The second relates to the reproduction

operator allowing to invade the search space and to

derivate new CUIs. Only the best CUIs are chosen

for the next population depending on the objective

functions. We highlighted three objective functions

based upon the usability sub-characteristics: learn-

ability, undrestandability and operability.

This research may be extended, built upon and

further elaborated as multiple studies may be under-

taken in this area. This involves, in particular, adding

more usability attribues to be optimized. Besides, to

further enhance the performance of our optimization

process, we propose to combine another optimization

algorithm with the evolutionary algorithm. Indeed,

this algorithm is not able to explore the whole search

space by restricting a set of solutions. It would there-

fore be promising to combine the evolutionary algo-

Optimizing the Usability of User Interfaces based on Multi-objective Evolutionary Algorithms

283

rithm with a local search method.

REFERENCES

Abrah

ao, S. and Insfran, E. (2017). Evaluating software

architecture evaluation methods: An internal replica-

tion. In Proceedings of the 21st International Confer-

ence on Evaluation and Assessment in Software Engi-

neering, pages 144–153. ACM.

Afzal, W., Torkar, R., and Feldt, R. (2009). A systematic

review of search-based testing for non-functional sys-

tem properties. Information and Software Technology,

51(6):957–976.

Ammar, L. B., Trabelsi, A., and Mahfoudhi, A. (2016).

A model-driven approach for usability engineering

of interactive systems. Software Quality Journal,

24(2):301–335.

Bastien, J. C. and Scapin, D. L. (1993). Ergonomic criteria

for the evaluation of human-computer interfaces. PhD

thesis, Inria.

Bavota, G., Di Penta, M., and Oliveto, R. (2014). Search

based software maintenance: Methods and tools. In

Evolving software systems, pages 103–137. Springer.

Bechikh, S., Said, L. B., and Gh

edira, K. (2011). Search-

ing for knee regions of the pareto front using mobile

reference points. Soft Computing, 15(9):1807–1823.

Boussa

ıd, I., Siarry, P., and Ahmed-Nacer, M. (2017).

A survey on search-based model-driven engineering.

Automated Software Engineering, 24(2):233–294.

Calvary, G., Coutaz, J., Thevenin, D., Limbourg, Q., Bouil-

lon, L., and Vanderdonckt, J. (2003). A unifying ref-

erence framework for multi-target user interfaces. In-

teracting with computers, 15(3):299–308.

Chanquoy, L., Tricot, A., and Sweller, J. (2007). La charge

cognitive: Th

eorie et applications. Armand Colin.

da Silva, A. R. (2015). Model-driven engineering: A survey

supported by the uniﬁed conceptual model. Computer

Languages, Systems & Structures, 43:139–155.

Fleck, M., Troya, J., Kessentini, M., Wimmer, M., and Alk-

hazi, B. (2017). Model transformation modulariza-

tion as a many-objective optimization problem. IEEE

Trans. Softw. Eng, 43(11):1009–1032.

Gajos, K. Z., Weld, D. S., and Wobbrock, J. O. (2010).

Automatically generating personalized user interfaces

with supple. Artiﬁcial Intelligence, 174(12):910–950.

Harman, M. and Jones, B. F. (2001). Search-based software

engineering. Information and software Technology,

43(14):833–839.

Hentati, M., Ammar, L. B., Trabelsi, A., and Mahfoudhi, A.

(2016a). An approach for incorporating the usability

optimization process into the model transformation. In

International Conference on Intelligent Systems De-

sign and Applications, pages 879–888. Springer.

Hentati, M., Ammar, L. B., Trabelsi, A., and Mahfoudhi,

A. (2016b). Model-driven engineering for optimizing

the usability of user interfaces. In ICEIS 2016 - Pro-

ceedings of the 18th International Conference on En-

terprise Information Systems, Volume 2, Rome, Italy,

April 25-28, 2016, pages 459–466.

Hentati, M., Trabelsi, A., Ammar, L. B., and Mahfoudhi, A.

(2018). Motuo: An approach for optimizing usability

within model transformations. Arabian Journal for

Science and Engineering, pages 1–17.

Hentati, M., Trabelsi, A., Benammar, L., and Mahfoudhi,

A. (2019). Search-based software engineering for

optimising usability of user interfaces within model

transformations. IET Software, 13(5):368–378.

niguez-Jarr

ın, C., Panach, J. I., and L

opez, O. P. (2020).

Improvement of usability in user interfaces for mas-

sive data analysis: an empirical study. Multimedia

Tools and Applications, pages 1–32.

ISO, I. (2005). Iec 25000 software and system engineering–

software product quality requirements and evaluation

(square)–guide to square. International Organization

for Standarization, pages 3–20.

Mkaouer, M. W., Kessentini, M., Shaout, A., Koligheu, P.,

Bechikh, S., Deb, K., and Ouni, A. (2020). Many-

objective software remodularization using nsga-iii.

arXiv preprint arXiv:2005.06510.

Ouni, A., Kessentini, M., Sahraoui, H., and Boukadoum, M.

(2013). Maintainability defects detection and correc-

tion: a multi-objective approach. Automated Software

Engineering, 20(1):47–79.

Pitangueira, A. M., Maciel, R. S. P., and Barros, M. (2015).

Software requirements selection and prioritization us-

ing sbse approaches: A systematic review and map-

ping of the literature. Journal of Systems and Soft-

ware, 103:267–280.

Rachmawati, L. and Srinivasan, D. (2009). Multiobjective

evolutionary algorithm with controllable focus on the

knees of the pareto front. IEEE Transactions on Evo-

lutionary Computation, 13(4):810–824.

Raneburger, D., Kaindl, H., and Popp, R. (2015). Strategies

for automated gui tailoring for multiple devices. In

System Sciences (HICSS), 2015 48th Hawaii Interna-

tional Conference on, pages 507–516. IEEE.

Seffah, A., Donyaee, M., Kline, R. B., and Padda,

H. K. (2006). Usability measurement and metrics:

A consolidated model. Software Quality Journal,

14(2):159–178.

ICSOFT 2021 - 16th International Conference on Software Technologies

284