Assessing the Quality of User-interface Modeling Languages
Francisco Morais
1
and Alberto Rodrigues da Silva
1,2
1
Instituto Superior Técnico, Universidade de Lisboa, Lisbon, Portugal
2
INESC— ID, Lisbon, Portugal
Keywords:
Appropriateness, Domain-specific Languages, Evaluation, Quality, User-interface Modeling Languages.
Abstract:
Model-Driven Development (MDD) is an approach that considers model as first citizen elements in the context
of software development. Since there are so many modeling languages, there is a need to compare them and
choose the best for each concrete situation. The selection of the most appropriate modeling language may
influence the output’s quality, whether it is only a set of models or software.
This paper introduces ARENA, a framework that allows to evaluate the quality and effectiveness of modeling
languages. Then we will apply ARENA to a specific subset of User-Interface Modeling Languages (namely
UMLi, UsiXML, XIS and XIS-Mobile), taking into account some of their characteristics and the influence
they have when models are generated.
1 INTRODUCTION
A modeling language might be classified as general-
purpose (GPML) or domain-specific modeling lan-
guage (DSML) (van Deursen et al., 2000; Luoma
et al., 2004; Mernik et al., 2005; Kosar et al., 2010).
A GPML is characterized by having a greater num-
ber of generic constructs, which encourages a wider
and widespread use in different fields of application.
UML or SysML are popular examples of GPMLs
by providing large sets of constructs and notations
used for specifying and documenting, respectively,
software systems according to the object-oriented
paradigm, or for system engineering. On the other
hand, DS(M)Ls tend to use few constructs or concepts
that are closer to its application domain.
Since a DS(M)L is expressed using domain con-
cepts, it is normally easier to read, understand, vali-
date and communicate with, facilitating cooperation
between developers and domain experts. Moreover,
some argue that DS(M)Ls can improve productiv-
ity, reliability, maintainability and portability (van
Deursen et al., 2000). However, the use of a DS(M)L
can raise some problems, such as the cost of learn-
ing, implementing and maintaining a new language,
as well as the support tools to develop with it (Mernik
et al., 2005).
Either for GPMLs as for DSMLs, nowadays there
is a great variety of modeling languages. For in-
stance, BPMN is specific to business process design
although UMLs activity diagram is also adequate for
this purpose (Object Management Group, 2013; Ob-
ject Management Group, 2011), DEMO is specific to
enterprise architecture (Dietz, 2001), XIS to Interac-
tive Applications, and Petri Nets to Distributed Sys-
tems (Desel and Juhás, 2001). Since there are more
languages and approaches than domains, this results
in overlapping effort for researchers and disorienta-
tion for modelers.
In order to help professionals selecting the most
suitable modeling language to work on so many spe-
cific and different contexts, evaluation frameworks
are needed. An evaluation framework is a set of prop-
erties, metrics, concepts and other parameters that
compare and assert the languages’ characteristics. For
this work, we have proposed ARENA Framework to
enlighten individuals that work with models on this
area. Our goal in this article is to help the user choos-
ing the most appropriate modeling language, in order
to assure a quality output, to feel more supported and
to work in a more rational way.
In this paper we propose a framework to evalu-
ate and compare the quality of User-Interface Model-
ing Languages (UIMLs), focused on either their gen-
eral and specific characteristics. The definition of a
modeling language involves multiple aspects or facets
which have to be taken into consideration, namely ab-
stract syntax, concrete syntax, and semantics.
UIMLs are DSMLs that are specifically used for
modeling the user interface of desktop, web or mo-
311
Morais F. and Rodrigues da Silva A..
Assessing the Quality of User-interface Modeling Languages.
DOI: 10.5220/0005369303110319
In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 311-319
ISBN: 978-989-758-097-0
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
bile applications, supporting the design and imple-
mentation phases (Achilleos et al., 2008). They de-
note conceptual abstractions that are present on user
interfaces (Frank, 1996). This paper analyses and
compares the main properties of four UIMLs, namely:
UMLi (Silva, 2002), UsiXML (Limbourg et al.,
2004), XIS (Silva et al., 2003) (Martins and Silva,
2007) (Silva et al., 2007) and XIS-Mobile (Ribeiro
and Silva, 2014b) (Ribeiro and Silva, 2014a). We
have selected them due to their support, complete
documentation, project visibility and availability of
their papers.
We are aware of other UIMLs such as Di-
aMODL (Trætteberg, 2008), IFML (Object Manage-
ment Group, 2014), MARIA XML (Paternó et al.,
2009) or WebML (Ceri et al., 2002). However, we
acknowledge that they don’t fit the domain and our
criteria as well as the previous four. WebML had
strong influence in IFML, when Object Management
Group created this language and that IFMLs most re-
cent version (from 2014) is still in Beta. Therefore,
we consider that there were too much similarities be-
tween both and comparing them would not prove ef-
fective. Regarding DiaMODL and MARIA XML, the
first one is too focused on Dialog Modeling, Dataflow
and State logic while the second is not quite appropri-
ate due to its models’ development being highly de-
pendable in Service-Oriented Architectures and Web
Services, which implies exploiting annotations at de-
sign time and the language itself at runtime to support
dynamic generation of user interfaces, thus limiting
usability and analysis.
This paper is organized in 6 sections: Section 1
gives a brief explanation and context of this work.
Section 2 introduces the background for this re-
search, that includes a definition of Quality and its
importance for this work, a brief overview of Mod-
eling Languages and an analysis of other Evalua-
tion Frameworks. Section 3 presents the proposed
ARENA Framework to the previously referred prob-
lem. Section 4 evaluates the quality of four UIMLs
with the ARENAs properties and metrics. Section 5
discusses the evaluation made based on the ARENA
Framework. Finally, section 6 concludes this paper.
2 BACKGROUND
This section introduces the background of this re-
search, namely Quality, Modeling Languages and
Evaluation Frameworks.
2.1 Quality
Regarding a model that may generate software, its
quality can be divided into internal and external (In-
ternational Organization for Standardization and In-
ternational Electrotechnical Commission, 2001).
Internal quality consists on the characteristics of
the software product from an internal view. Inter-
nal quality requirements are used to specify properties
of interim products. They can include static and dy-
namic models, other documents and source code (In-
ternational Organization for Standardization and In-
ternational Electrotechnical Commission, 2001).
External quality is defined by the characteristics of
the software product from an external view. It is the
quality of the executed software, which is typically
measured and evaluated while testing in a simulated
environment with simulated data using external met-
rics. These metrics should be aligned with the exter-
nal quality requirements, which goal is to specify the
required level of quality from the external view (In-
ternational Organization for Standardization and In-
ternational Electrotechnical Commission, 2001).
Quality is not yet a common priority when devel-
oping a Domain-Specific Language (DSL). The fo-
cus is currently on getting up a systematic develop-
ment of these Modeling Languages, a goal that has
not yet been reached because it has been doing more
study of the technical aspects of DSLs’ design and
implementation, such as: case studies and technical
reports on individual DSLs; design approaches and
techniques for implementing DSLs; and integrating
DSLs with other developmental approaches (Strem-
beck and Zdun, 2009).
2.2 Modeling Languages
A modeling language (ML) is a set of words and
symbols, supported by validation rules and semantics,
which make it possible to create models or diagrams.
It can be graphical/visual or textual, depending on its
scope and domain (He et al., 2007).
A modeling language is created from a metamodel
within a certain domain, therefore, it can be seen as
a model of a modeling language (Ma et al., 2004).
In order to give the modeling languages functional-
ity to use and evaluate the models, different types of
mechanisms are included in the metamodeling pro-
cess (algorithms, generic mechanisms, specific mech-
anisms and hybrid mechanisms) (Karagiannis and
Kühn, 2002). In other words, the metamodel is able to
highlight the properties of the model, derived from its
capacity of abstraction. There are several metamod-
eling approaches. The most commonly used is Meta
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
312
Object Facility (MOF) and it has been used, for in-
stance, in the development of UML and SysML.
In terms of representation type, MLs can be di-
vided in two groups: graphical and textual. On one
hand, graphical MLs use a diagram technique with
named symbols that represent concepts and lines that
connect the symbols and represent relationships and
various other notation to represent constraints. On the
other hand, textual MLs use standardized keywords
accompanied by parameters or natural language terms
and phrases to make computer-processable expres-
sions. Considering modeling approaches, in the field
of computer science, more specific types of MLs
have recently emerged, due to new challenges on
very different areas, namely: Algebraic, Behavioural,
Discipline-Specific, Domain-Specific, Framework-
Specific, Object-Oriented or Virtual Reality.
An ML is very important because it helps to elu-
cidate their stakeholders that have modeling expertise
to understand what and how they can make a repre-
sentation of the system-of-interested. Also, its fea-
tures such as syntax, abstraction and compatibility
with programs can facilitate its use and its correspon-
dent metamodels may give a perception about the ad-
equacy of the language to the concrete situation.
2.3 Evaluation Frameworks
Hereby, we are exposing three frameworks that have
been developed to evaluate the quality of concep-
tual models and three business process modeling lan-
guages.
SEQUAL is a framework that was presented in
1994 with the goal of evaluating systematically the
quality of information systems and other conceptual
models through the notions of syntactic, semantic and
pragmatic applied to them (Krogstie et al., 1995). One
year later it was extended to more three features, in-
spired by FRISCO’s six semiotic layers of communi-
cation. It is considered the first framework to have the
objective on evaluating models’ quality.
Essentially, the first version of the framework only
had four concepts Model, Domain, Language and
Audience interpretation and six pair relations be-
tween them. Its extension, in 1995, has brought
the addition of a new entity (Participant knowledge)
and the creation of four relationship types. In 2012,
Krogstie makes a third instance of his framework
and adds "deontic" as a new quality criteria that fo-
cuses on what is right and needed for the organiza-
tion’s goals (Krogstie, 2012). SEQUALs formulas
are based on the audience, the language, the model,
the domain, the audience’s knowledge and the audi-
ence’s interpretation.
On 2005, a paper about focusing the standardiza-
tion of business process modeling was presented. It
intended to compare some Modeling Languages that
could be used for this context. The three selected
BPM languages were EEML, UML and BPMN,
and their framework comprised the following items:
Goals of modeling task, language extension, domain,
externalized model, knowledge of the stakeholders,
the social actors’ interpretation and the technical ac-
tors’ interpretation (Nysetvold and Krogstie, 2005).
On 1997, Teeuw and van der Berg published a pa-
per about their perspective on general quality crite-
ria for conceptual models and a framework that was
used to evaluate the redesign of business processes.
Like SEQUAL, they also defend that a good, qual-
ity model must have syntactic, semantic and prag-
matic qualities and, considering that the concepts
can be captured with a suitable language (assuring
the first quality), they present as criteria for the sec-
ond and the third: completeness, inheritance, clarity,
consistency, orthogonality and generality (Teeuw and
van der Berg, 1997). Considering the Testbed evalu-
ation framework, it was applied to behaviour models
and it was designed with 4 dimensions: Functionality,
Ease of use, Business Process Redesign (BPR) trajec-
tory and General.
Although they use interesting criteria, the re-
lated work frameworks don’t satisfy the goal of this
work because they evaluate other matters instead of
UIMLs, respectively, the Quality of Conceptual Mod-
els, the Quality of Business Process and Business Pro-
cess redesign.
The selection of the appropriate Modeling Lan-
guage influences the quality of the final output,
whether it is a model or a software application (In-
ternational Organization for Standardization and
International Electrotechnical Commission, 2008;
Krogstie et al., 1995; Rech and Bunse, 2009). Qual-
ity integration in software development processes that
use models, such as Model-Driven Software Develop-
ment, is the ultimate reason that led us to propose this
framework. ARENA displays a Framework that, after
analysing the MLs’ characteristics, shows the evalua-
tion of each one and helps the user to choose the most
appropriate ML, in order to assure a quality output.
It is the most appropriate evaluation framework for
modelling languages, because it is not only divided
into general and specific properties (which takes into
account the considered domain), but also because it is
an extensive comparison and assessment, resembling
the CMS Matrix.
AssessingtheQualityofUser-interfaceModelingLanguages
313
ML_Quality
Property
Metric
Dimension
ML_Info
General_Property Specific_Property
Range
«enumeration»
NotationKind
1 1
11
0..*
1..*
0..*
1..*
uses
1..*
1
Figure 1: ARENA Framework — Main Concepts.
3 THE ARENA FRAMEWORK
Figure 1 shows that a Modeling Language’s Quality
is the central class. It contains the final quantitative
output. This class uses the Modeling Language’s
information as an input and has a Range.
This one is calculated using a formula that
receives as an input the values of the quality and
quantity evaluations (respectively represented as
Dimension and Metric classes) and multiplies each
by a previously defined weight. The returned output
is a value that represents the language’s Quality
within a rating scale, as explained hereinafter on
section 4. All domains share the General Properties,
as opposite to Specific Properties, that differ from
each other (e.g. UIMLs, BPMLs etc). The Notation
Kind enumeration is developed as a droptext, so the
user can select one option from the displayed list.
The proposed framework intends to bridge that gap
and it is composed by general and specific properties.
3.1 General Properties
Interoperability. The language must be fully com-
patible with several tools, i.e. should allow to do
the same tasks and diagrams in different software
tools. Also, it shall be possible to combine with
another modeling and programming languages (PLs)
and tools, being supported by mechanisms that guar-
antee those possibilities. Dimensions and Metrics:
Number of Compatible Applications (International
Organization for Standardization and International
Electrotechnical Commission, 2008), Number of In-
tegration Mechanisms (Karagiannis and Kühn, 2002)
and Tool Supportability.
Notation. A Notation or concrete syntax is a set
of signs that enable to represent models. A model-
ing language may have two types of notation: graph-
ical/visual or textual. Dimensions: Representation
Type and Supporting Mechanisms.
Size. Completeness is one of the biggest chal-
lenges regarding the development of modeling lan-
guages and respective models (Baader et al., 2003;
Hoppenbrouwers et al., 2005; Teeuw and van der
Berg, 1997). This property states that the language’s
meta-model shall include the most important con-
cepts. Therefore, the meta-modeling approach is es-
sential for assuring that the modeling languages allow
producing concrete quality models. Metrics: Number
of Views, Number of Classifiers and Number of Re-
lationships.
3.2 Specific Properties
Application Actions. This property lists a series of
actions that the generated application can support, in
different contexts.
Other Features. It is a set of properties of
the UIMLs that provide additional information about
them. In this paper, they will be considered, but not
weighted to calculate the quality.
Pattern Usage. This property displays a list of
simple and reusable templates that help the user solv-
ing common problems that appear during the design
phase.
Tool Support - Model to Model (M2M) Trans-
formations. This property tells if the compatible
programs are capable of transforming the language’s
models format into another format or if the language
can produce more than one format.
Tool Support - Model to Text (M2T) Transfor-
mations. This property is a key aspect of model-
driven development. It refers the language’s ability
or inability to generate textual artefacts from models
and if so, which mechanisms or techniques make it
possible.
Tool Support - Validation. This property has
the purpose of listing which systems, applications or
mechanisms can analyse the models created by the
user and validate them.
User Interactions. This property intends to show
which ways users can interact with the application,
whether it is with mouse, keyboard, touch or other
means.
Widget Types. This property lists a set of graph-
ical user interface elements (either structural or be-
havioural) that the language makes available for the
designer.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
314
Table 1: UIMLs comparison based on the ARENA Framework.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
Property
Language
Property
Group
UMLi UsiXML XIS XIS-Mobile
Average
(UIMLs)
Number of Views Abstract Syntax 4 5 6 6 5
Number of Classifiers " 13 12 18 46 22
Number of Relationships " 11 20 12 16 15
Representation Type
Concrete
Syntax
Graphical Textual Graphical Graphical
Supporting Mechanisms " Core meta-modelling
XML Metamodel and
Cameleon reference
framework
UML Profile UML Profile
Tool Supportability " ARGOi
Eclipse, Visual Studio,
Notepad++, Kate,
W3schools.com and others
ProjectIT Studio Enterprise Architect
Application Actions
Specific
Properties
OK, Cancel, Search, Back,
Next, Up, Down, Quit,
other customizable actions
(N/A)
CRUD, OK, Cancel,
Navigate, Select, Close,
Associate, Dissociate,
other customizable actions
CRUD, OK, Cancel,
Delete All, Open Browser,
Web Service, Navigate,
Select, other customizable
actions
>8
User Interactions "
Click, Select, (Keyboard)
Type, Scroll
Move pointer, Click,
Double click, Depress,
Release, Drag over, Drag
drop, Focus, Select,
Choose, Toggle, View
Click, Select, Drag,
(Keyboard) Type
Tap, Double Tap, Long
Tap, Swipe, Pinch, Stretch,
(Touchscreen) Type
7
Pattern Usage "
Abstract Presentation
Pattern, Concrete
Interaction Object
Concepts & Task Model,
Abstract UI, Concrete UI,
Final UI, Inter-model
mapping, Context
translation
Composite, Single Choice,
Multiple Choice, List
Selection, Continuous
Filter, Menu Navigation,
Grid Layout, Tab Menu,
Tabular Set, Double List
Composite, Single Choice,
Multiple Choice, List
Selection, Single Text
Entry, Multiple Text Entry,
Springboard, List Menu,
Tab Menu, Option Menu
and 10 more
10
Tool Support - Validation "
ARGOi validates both
UML and UMLi models,
due to both grammars are
specified in terms of the
UML metamodel
(N/A)
Eclipse .NET, .NET
Framework and Project
IT’s three components:
Requirements, UML
Modeler and MDD Code
Generator
XIS-Mobile Framework,
namely EA’s Model
Validator, supports model
validation
Tool Support - M2M
Transformations
" No (N/A) No
Yes, using Enterprise
Architect’s MDG
technologies
Tool Support - M2T
Transformations
" No (N/A)
Yes, defining architectures,
templates, and interface
generation processes that
are compatible with
Windows Forms.NET and
ASP.NET platforms
Yes, using XMI-validated
and generated models,
Acceleo and the OS’s
compatible programming
languages. It can generate
Java, C#, Objective-C,
XML and XAML
Number of Compatible
Applications
" 1 (ArgoUML) 1 (Eclipse) 1 (ProjectIT Studio) 1 (Enterprise Architect) 1
Number of Integration
Mechanisms
"
3 (ARGOi, OCL rules and
LOTOS rules)
4 (MDG Technologies,
XML Parser, UsiGesture
and UsiDistrib)
2 (Eclipse .NET and .NET
Framework)
2 (MDG Technologies,
XML Parser)
3
Widget Types "
Label, Text field, Combo
box, Selectable list, Button
Push button, List box,
Check box, Window,
Panel, Table, Cell, Dialog
box, Embeded multimedia,
Menu, Spin button
Button, Text box, List,
Menu, Window, Link,
Search bar, Checkbox,
Radius button, Drop-down
list, Form, Dialog, Label
and 15 more
Button, Text box, List,
Menu, Window, Link,
Search bar, Checkbox,
Radius button, Drop-down
list, Label, Image, Date
Picker and 9 more
17
Other Features "
It is possible to model
Activity Diagrams in
UMLi, using the Use Case
Diagram and the
InitialInteraction construct.
It implements the µ7
concept, as it is Device-,
User-, Culturality-,
Organization-, Context-,
Modality- and
Platform-Independent.
Supports Windows
Forms.NET, ASP.NET and
JSP, using Model-to-Text
transformations.
Model validation uses a set
of rules defined in C#,
implemented with EA’s
Automation Interface. It is
possible to generate
User-Interfaces View
models on EA.
4 APPLYING ARENA TO
EVALUATE USER-INTERFACE
LANGUAGES
Table 1 summarizes the main properties of four anal-
ysed UIMLs UMLi (Silva, 2002), UsiXML (Lim-
bourg et al., 2004), XIS (Silva et al., 2003) (Mar-
tins and Silva, 2007) (Silva et al., 2007) and XIS-
Mobile (Ribeiro and Silva, 2014b) (Ribeiro and Silva,
2014a). The columns Number of Views, Number of
Classes and Number of Relationships represent the
metrics with the same name, from ARENAs property
Size. The ARENA property Notation is represented
by two dimensions: Representation Type and Sup-
porting Mechanisms. The column Tool Supportability
refers to a dimension from ARENAs property Inter-
operability, as well as metrics Number of Compatible
Applications and Number of Integration Mechanisms.
The remaining columns have exactly the same name
and meaning of the specific properties, as shown on
the previous section. The quality value of each lan-
guage is determined by the formula below, designated
as the ARENA General Equation of Quality.
AssessingtheQualityofUser-interfaceModelingLanguages
315
QualityLanguage
n
=
j
i=1
w
i
r
i
=
= w_absStxProps r_absStxProps + w_concStxProps r_concStxProps + w_specProps r_specProps =
= w_numViews r_numViews + w_numClassi f s r_numClassi f s +w_numRelats r_numRelats+
+w_repType r_repType + w_suppMechs r_suppMechs + w_toolSupp r_toolSupp+
+w_numAppActs r_numAppActs + w_numUserInts r_numUserInts + w_numPatterns r_numPatterns+
+w_Valid r_Valid + w_M2MTr f s r_M2MTr f s + w_M2T Tr f s r_M2T Tr f s+
+w_numCompApps r_numCompApps +w_numIntMechs r_numIntMechs+ w_numWidgets r_numWidgets
The notation used on the formula above has
the following meaning: i Framework’s met-
ric/dimension; j Sum of the number of metrics
with the number of dimensions; n Language’s
name; r Metric’s/dimension’s rate; w Met-
ric’s/dimension’s weight. The rating each property
can have is the following: 1 — Very Low; 2 — Low;
3 — Medium; 4 — High; 5 — Very High.
If quantifiable and with equal level of importance,
each property’s value is compared towards that
property’s average value in this work. Its rating is
given according to how far it is from the average,
upwards or downwards, according to the follow-
ing ranges: Value < AvgValue 1/2 AvgValue ;
Value [AvgValue 1/2 AvgValue,AvgValue
1/4 AvgValue[ ; Value [AvgValue 1/4
AvgValue,AvgValue + 1/4 AvgValue] ; Value
]AvgValue + 1/4 AvgValue, AvgValue + 1/2
AvgValue] and Value > AvgValue + 1/2 AvgValue.
The exceptions (non-quantifiable properties) are:
Representation Type The most popular value has
higher rating than the least; Supporting Mechanisms
Since it’s very specific for each language, this
rating reflects easiness to understand and extend
the metamodel; Tool Supportability It may be
unique for each language, so its rating is focused on
the programs’ general usability and functionality;
and Tool Support (Validation, Model to Model and
Model to Text) Since the programs referred on
the Tool Supportability property generate models, we
assumed that Validation is correctly done, as well as
the transformations.
The languages in which we don’t know how these
characteristics are covered were assigned 1 (Very
Low) and the languages that don’t have these features
were assigned 2 (Low). Each rated property will be
multiplied by its weight, as referred on section 3, and
the sum of this products will return the language’s
quality, according to the ARENA General Equation
of Quality.
5 DISCUSSION
Having four UIMLs compared, it is possible to see
that, for the same domain, the languages do not differ
much in terms of number of views and number of rela-
tionships, but they do when it comes about supporting
mechanisms and tool supportability. This can be ex-
plained by their maturity and popularity. Also, there
are patterns concerning number of classifiers and no-
tation, both with a single exception.
If the Number of Views seems to be balanced for
these four languages (all have between 4 and 6), it
is not possible to conclude the same about the Num-
ber of Classifiers, because UsiXML, UMLi and XIS
have, respectively, 12, 13 and 18 and XIS-Mobile has
46, resulting in a difference of 34 items to the lowest
number. Less unbalanced is the Number of Relation-
ships — it varies from UMLi’s 11 to UsiXMLs 20.
For Representation Type, clearly there is a prefer-
ence for Graphical in disfavour of Textual, probably
due to easiness to see the models in a clearer way and
according to the WYSIWYG paradigm. About Sup-
porting Mechanisms, as expected due to their nature,
XIS and XIS-Mobile have the same value, oppos-
ing to UMLi that is an extension to UML and there-
fore uses a simpler approach and UsiXML that, be-
ing based on XML, results in a different architecture
that is complemented by Cameleon. Considering Tool
Supportability, all languages follow different tenden-
cies, with a clear advantage to UsiXML. This might
be due to XMLs universality (as it is used in RSS,
SOAP messages and XHTML), it is both human- and
machine-readable and it is capable of representing all
Unicode characters.
XIS is the top language concerning to Application
Actions, since it allows the software to do at least
9 different actions. Contrasting with that, UsiXML
is the better one in terms of User Interactions, since
it provides 12 gestures that can be done, while XIS
and UMLi only have 4 defined. The language that in-
cludes more Patterns is XIS-Mobile, Composite being
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
316
Table 2: UIMLs evaluated with ARENA Framework.
`
`
`
`
`
`
`
`
`
`
`
`
`
`
`
Property
Language
Weight UMLi UsiXML XIS XIS-Mobile
Number of Views 0.033 3 3 3 3
Number of Classifiers 0.033 2 2 3 5
Number of Relationships 0.033 2 4 3 3
Representation Type 0.033 4 3 4 4
Supporting Mechanisms 0.033 3 3 3 3
Tool Supportability 0.033 3 4 2 4
Application Actions 0.1 3 1 3 3
User Interactions 0.1 2 5 2 3
Pattern Usage 0.1 1 2 3 5
Tool Support - Validation 0.1 3 1 3 3
Tool Support - M2M
Transformations
0.1 2 1 2 4
Tool Support - M2T
Transformations
0.1 2 1 4 4
Number of Compatible
Applications
0.05 3 3 3 3
Number of Integration
Mechanisms
0.05 3 3 3 3
Widget Types 0.1 1 2 5 4
Other Features
Total Quality 1 2 2 3 4
the only one of Structural Design type and the remain-
ing 19 of UI Design, that is 10 more than XIS. Still
about this property, UsiXML has 4 which are based
on Cameleon reference framework and UMLi has 2
patterns, each for a UML package that was extended.
About Tool Support - Validation, each language
has its own method, always influenced by the compat-
ible applications and the ones shown in Tool Support-
ability column, which in many cases, are the same.
We don’t know how it’s done for UsiXML, but for
XIS it is the most extensive process. Now referring
to Tool Support - Model to Model Transformations,
only XIS-Mobile has this feature implemented, tak-
ing advantage of EAs capabilities, while regarding
Tool Support - Model to Text Transformations, XIS
and XIS-Mobile provide this MDD-core feature, as
opposed to UMLi, with a great focus on Web tech-
nologies. Again, we know nothing about these two
properties on the UsiXML case.
Now regarding the Number of Compatible Ap-
plications, all languages have the same value, there-
fore we considered than none stood out more than the
other. It seems to be a bit contradictory to the Tool
Supportability property. The difference is that the lat-
ter doesn’t take into account the language’s suitable
program but the produced models’ compatibility with
other platforms. For the Number of Integration Mech-
anisms, despite some are easier to work with (XML
Parser) than others (LOTOS rules), all languages have
received the same rating, due to the number is close
to the average of this property, in this work.
The attractive and useful Widget Types inform
that XIS and XIS-Mobile have a higher number of
items (28 and 22, accordingly), although not exactly
the same. UsiXML stands with 11 and UMLi has only
5, which is expected from an extension that claims to
"be a conservative extension of UML" and "should
introduce as few new models and constructs into the
UML as possible" (Silva, 2002).
According to Table 2, for this evaluation, XIS-
Mobile has the highest quality value 4 among
these four UIMLs, in the considered domain.
6 CONCLUSION
This paper presents an overview of academic research
related to the problematic of choosing the best from
several modeling languages and the quality assess-
ment frameworks as a response to that problem. We
have evaluated four UIMLs using ARENA, a frame-
work oriented to that solution. In this case, ARENAs
most adequate dimensions and metrics were used,
along with User-Interface specific properties, defined
for this purpose and context. With this analysis, it
is possible to list the features in which each UIML
stands out from the others, either because they are bet-
ter or they are unique. Clearly there is a tendency to
bet on graphical notations. This can be due to guide-
lines that compose the well known Model-Driven En-
AssessingtheQualityofUser-interfaceModelingLanguages
317
gineering. Also an advantage, is to use these graph-
ical models to implement the alternative approach to
produce software: Model-Driven Software Develop-
ment.
The metamodel types chosen vary very much from
modeling language to modeling language, since some
use MOF, others BNF or UML Profiles, and even
MVC is still used. Another possible choice is to ex-
tend other MLs metamodel. Tool support is very im-
portant, in terms of usability and functionality. They
are also responsible to render and validate the pro-
duced models, so this aspect can be the quality bot-
tleneck. In terms of look-and-feel, it is very impor-
tant for the designer to have available design patterns
and builder tools, so he can produce an attractive and
easy interface. He must also be able to choose be-
tween several actions and widgets, so the mobile, web
or desktop software application can have an attractive
layout and great impact.
In the future, ARENA will also be used to com-
pare and evaluate General-Purpose Modeling Lan-
guages, such as UML and SysML or Business Process
Modeling Languages, namely BPMN and DEMO, al-
ways taking into account the general and the domain
specific characteristics.
ACKNOWLEDGEMENTS
This work was partially supported by the ARENA
2012 IBM Country Project, and by national funds
through Fundação para a Ciência e a Tecnologia
(FCT) with references UID/CEC/50021/2013 and
EXCL/EEI- ESS/0257/2012 (DataStorm).
REFERENCES
Achilleos, A., Yang, K., Georgalas, N., and Azmoodech,
M. (2008). Pervasive service creation using a model
driven petri net based approach. In Wireless Commu-
nications and Mobile Computing Conference, 2008.
IWCMC ’08. International, pages 309–314.
Baader, F., Calvanese, D., McGuiness, D., Nardi, D., and
Patel-Schneider, P. (2003). The Description Logic
Handbook: Theory, Implementation and Applications.
Cambridge University Press.
Ceri, S., Fraternali, P., Bongio, A., Brambilla, M., Comai,
S., and Matera, M. (2002). Designing Data-Intensive
Web Applications. Morgan Kaufmann Publishers Inc.,
San Francisco, CA, USA.
Desel, J. and Juhás, G. (2001). What Is a Petri Net?, volume
2128 of LNCS. Springer-Verlag.
Dietz, J. (2001). DEMO: Towards a discipline of organi-
sation engineering. European Journal of Operational
Research, 128:351–363.
Frank, M. R. (1996). Model-based User Interface Design by
Demonstration and by Interview. College of Comput-
ing, Georgia Institute of Technology 1996. Directed
by James Foley.
He, X., Ma, Z., Shao, W., and Li, G. (2007). A Metamodel
for the Notation of Graphical Modeling Languages.
In Proceedings of the 31st Annual International Com-
puter Software and Applications Conference - Volume
01, volume 1 of COMPSAC ’07, pages 219–224. IEEE
Computer Society.
Hoppenbrouwers, S., Proper, E., and van der Weide, T. P.
(2005). A Fundamental View on the Process of Con-
ceptual Modeling. In Conceptual Modeling - ER 2005,
volume 3716, pages 128–143. Springer-Verlag.
International Organization for Standardization and Interna-
tional Electrotechnical Commission (2001). ISO/IEC
9126-1:2001(E) Quality Model.
International Organization for Standardization and Interna-
tional Electrotechnical Commission (2008). ISO/IEC
CD 25010.2 Software and Quality in use models.
Karagiannis, D. and Kühn, H. (2002). Metamodelling Plat-
forms. In Proceedings of the Third International Con-
ference EC-Web 2002 - Dexa 2002, volume 2455,
pages 182–195. Springer-Verlag.
Kosar, T., Oliveira, N., Mernik, M., Pereira, M. J. V.,
ˇ
Cre-
pinšek, M., da Cruz, D., and Henriques, P. R. (2010).
Comparing general-purpose and domain-specific lan-
guages: An empirical study. In Computer Science and
Information Systems, volume 7. University of Novi
Sad, Serbia.
Krogstie, J. (2012). Model-Based Development and Evo-
lution of Information Systems: A Quality Approach,
chapter 5, pages 249–280. Springer-Verlag London.
Krogstie, J., Lindland, O. I., and Sindre, G. (1995). Defin-
ing quality aspects for conceptual models. Faculty of
Electrical Engineering and Computer Science -, The
Norwegian Institute of Technology.
Limbourg, Q., Vanderdonckt, J., Michotte, B., Bouillon,
L., Florins, M., and Trevisan, D. (2004). USIXML:
A User Interface Description Language for Context-
Sensitive User Interfaces.
Luoma, J., Kelly, S., and Tolvanen, J.-P. (2004). Defin-
ing domain-specific modeling languages: Collected
experiences. In OOPSLA 4th Workshop on Domain-
Specific Modeling. ACM.
Ma, H., Shao, W., Zhang, L., Ma, Z., and Jiang, Y. (2004).
Applying OO Metrics to Assess UML Meta-models.
In UML 2004, LNCS, volume 3273, pages 12–26.
Springer-Verlag.
Martins, C. and Silva, A. R. (2007). Modeling user inter-
faces with the xis uml profile. In Proceedings of the
ICEIS 2007.
Mernik, M., Heering, J., and Sloane, A. M. (2005). When
and how to develop domain-specific languages. ACM
Computing Surveys, 37:316–344.
Nysetvold, A. G. and Krogstie, J. (2005). Assessing busi-
ness processing modeling languages using a generic
quality framework. Norwegian University of Science
and Technology.
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
318
Object Management Group (2011). OMG Unified Mod-
elling Language (OMG UML), Infrastructure - Ver-
sion 2.4.1. Object Management Group.
Object Management Group (2013). Business Process
Model and Notation (BPMN) - Version 2.0.2. Object
Management Group.
Object Management Group (2014). Interaction Flow Mod-
elling Language (IFML) - FTF - Beta 2 - Revision 21.
Object Management Group.
Paternó, F., Santoro, C., and Spano, L. D. (2009). MARIA:
A Universal, Declarative, Multiple, Abstraction-
Level Language for Service-Oriented Applications
in Ubiquitous Environments. ACM Transactions on
Computer-Human Interaction, 16:1–30.
Rech, J. and Bunse, C. (2009). Model-Driven Software De-
velopment: Integrating Quality Assurance. Informa-
tion Science Reference.
Ribeiro, A. and Silva, A. R. (2014a). Evaluation of xis-
mobile, a domain specific language for mobile appli-
cation development. In Journal of Software Engineer-
ing and Applications, number 7 in 11. Scientific Re-
search Publishing.
Ribeiro, A. and Silva, A. R. (2014b). Xis-mobile: A dsl
for mobile applications. In Proceedings of ACM SAC
2014 Conference. ACM.
Silva, A. R., de Sousa Saraiva, J., Silva, R., and Martins,
C. (2007). XIS - UML Profile for eXtreme Modelling
Interactive Systems. In Proceedings of the MOMPES
2007. IEEE Computer Society.
Silva, A. R., Lemos, G., Matias, T., and Costa, M. (2003).
The XIS Generative Programming Techniques. In
Proceedings of the 27th COMPSAC Conference. IEEE
Computer Society.
Silva, P. P. (2002). Object Modelling of Interactive Sys-
tems: The UMLi Approach. PhD thesis, University of
Manchester.
Strembeck, M. and Zdun, U. (2009). An approach for
the systematic development of domain-specific lan-
guages. Software Practice and Experience, 39:1253–
1292.
Teeuw, W. B. and van der Berg, H. (1997). On the Quality of
Conceptual Models. In 16th International Conference
on Conceptual Modeling - ER’97, volume 1331, pages
1–18. Springer-Verlag.
Trætteberg, H. (2008). Integrating Dialog Modeling and
Domain Modeling - the Case of Diamodl and the
Eclipse Modeling Framework. In Journal of Univer-
sal Computer Science, volume 14, pages 3265–3278.
J.UCS.
van Deursen, A., Klint, P., and Visser, J. (2000). Domain-
specific languages: An annotated bibliography. ACM
SIGPLAN Notices, 35(6):26–36.
AssessingtheQualityofUser-interfaceModelingLanguages
319