Connecting Feature Models and Ontologies in Software Product Lines

Gabriella Castro Barbosa Costa, Regina Braga, Jos

e Maria N. David and Fernanda Campos

Department of Computer Science, UFJF, Juiz de Fora, Brazil

Keywords:

Product Line, Feature Model, Ontology, Domain Engineering.

Abstract:

Feature models have been used as a way to describe the variability in Software Product Lines (SPL). Using this

model we can show variation points, variants and restrictions between variation points and variants. However,

restrictions and relations that can be expressed using this domain modelling technique are unsatisfactory.

Our proposal aims to solve this challenge connecting feature models and ontologies as domain modelling

technique, which can be used to provide additional information and can also add more semantics in the context

of Software Product Lines.

1 INTRODUCTION

A key issue in Software Product Line Engineering

(SPLE) is the speciﬁcation of the domain model,

which should express the variability of the SPL. To

do this, we can use domain models like feature mod-

els and ontologies, for example. Using feature models

we can show variation points, variants and restrictions

between variation points and variants. However, re-

strictions and relations that can be expressed using

this technique of domain modelling are unsatisfac-

tory, considering the lack of semantics between the

entities which composes the model as well as some

restrictions associated with the terms.

The purpose of this paper is to present two do-

main modeling techniques: feature modeling and on-

tology modeling which could be combined to obtain

the advantages of both and, as a result, to improve the

SPL domain speciﬁcation. In this work, we explore

the ontologies to get the main restrictions that need to

be applied to our SPL. These restrictions are used to

improve the choice of features related to the product

that will be developed. On the other hand, the feature

model is used to facilitate the understanding about the

features that need or not to be selected according to

the product that will be developed.

In Section 2, we describe the essence of feature

models and ontologies. In Section 3, three studies

that use feature modeling, ontologies or both of them

in SPLs are presented: (Czarnecki et al., 2006), (Jo-

hansen et al., 2010) and (Asikainen et al., 2007). In

the next section, our approach to connecting feature

models and ontologies in SPL is explained and an ex-

ample is shown. Conclusions about this work are pre-

sented in last section.

2 ESSENCE OF FEATURE

MODELS AND ONTOLOGIES

Feature models show the variation points and the vari-

ants of a SPL. This model should also include the re-

strictions between the variation points and the vari-

ants, as a variation point (or a variant) may require or

exclude a variation point (or a variant). However, fea-

ture models do not have a high power of description as

ontologies. For example, using only features models

there is no way to know when a particular feature is

more suitable to be chosen than another. This lack of

semantics in features models can be supplied through

the possibility of inference offered by ontologies.

In ﬁgure 1, when can see this lack of semantics of

a feature model. For example: using a feature model

we can make a restriction that shows that a ’pair-

wise local aligning’ is performed by an algorithm

called BLAST, but we can not express which BLAST

program is more appropriate for a particular type of

’pairwise local aligning’. Still according to ﬁgure 1,

we can see that the task ’multiple global aligning’

can be performed by two differents algoritms. But,

only through this feature model, we can not express

when the algorithm ’MultiAlign’ is more suitable than

the ’CLUSTAL W’ algorithm.

Ontology modeling is another domain modeling

technique. Ontology is ”a formal and explicit speciﬁ-

160

Castro Barbosa Costa G., Braga R., Maria N. David J. and Campos F..

Connecting Feature Models and Ontologies in Software Product Lines.

DOI: 10.5220/0003999101600163

In Proceedings of the 14th International Conference on Enterprise Information Systems (ICEIS-2012), pages 160-163

ISBN: 978-989-8565-11-2

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

Figure 1: The lack of semantics of a feature model.

cation of a shared conceptualization” (Gruber, 1993).

Assuming that ontologies can be used to model a

speciﬁc domain, it can also improve feature modeling

by providing additional and relevant information for

the domain of the SPL in which a feature model is

constructed. On the other hand, we need the feature

models because they are a simple way to understand

and to represent the features in the SPL.

3 RELATED WORK

(Czarnecki et al., 2006) explore the relationship be-

tween feature models and ontologies and suggest the

addition of extensions to the feature model in order

to move it closer to richer formalisms for specifying

ontologies. (Czarnecki et al., 2006) also explores the

idea of the feature models as views on ontologies. In

order to solve the gap between feature models and on-

tologies, our work suggests the combined use of fea-

ture models and ontologies. Differently from (Czar-

necki et al., 2006), we do not show how to improve

the feature model adding extensions on it. We also do

not use ontologies to align the view between the fea-

ture models or outlines of requirements. Our focus is

to use a feature model to represent the variability of

the SPL and the ontology to provide the restrictions

about the SPL that could not be represented only by

the features models. Moreover, we propose the use of

ontology and the possibility of making inferences in

it to improve the choices in the SPL.

(Johansen et al., 2010) explore the synergies be-

tween feature models and ontologies, showing map-

ping mechanisms that a user can establish and take

advantage of the relationship between feature models

and ontologies. They explore the gap between these

models and show scenarios where ontologies can be

applied for feature modeling. Our proposal also sug-

gests the use of ontologies and feature models to solve

the gap between these models, like (Johansen et al.,

2010). Differently, we focus on mapping the ontol-

ogy classes and features of the feature model, getting

the restrictions about the domain using the ontology.

After that, we show the features that need to be se-

lected on the feature model, at runtime.

In (Asikainen et al., 2007), a domain ontology for

modeling the variability in SPL, called Kumbang was

proposed. This ontology uniﬁes the feature model-

ing and the architecture modeling in SPL. Although

(Asikainen et al., 2007) uses ontology to represent the

features and the architecture of SPL, it should be con-

sidered that Kumbang ontology was described using

UML. As a result, the power of the semantics descrip-

tion of this language is not the same as OWL lan-

guages, considering also the inference mechanisms

that can be used in conjunction with OWL. In our

work, we use both a domain ontology (expressed us-

ing OWL-DL) and a feature model to support domain

modeling in a SPL.

4 CONNECTING FEATURE

MODELS AND ONTOLOGIES

We propose the development of a SPL to support sci-

entiﬁc applications construction in Biology, in order

to tackle the sequence alignment process problem.

We choose this domain to apply the proposed ap-

proach because our group has previous works which

involves bioinformatics, scientiﬁc software and use

of ontologies, as can be seen in (Matos et al., 2010),

(Mendes et al., 2011) and (Silva et al., 2012).

The initial architecture of the SPL is being de-

tailed considering its main modules (called man-

agers), as can be seen in Figure 2: (i) the Product

Line Manager, which interacts with the user and the

other managers in the architecture, coordinating their

behavior, (ii) the Domain Model Manager (which

includes the Variability Manager) is responsible for

maintaining the domain models in accordance with

the SPL scope, (iii) the Variability Manager, which

is responsible for managing the models that repre-

sent the variability in the SPL and (iv) the Archi-

tectural Manager is responsible for the SPL architec-

ture. These managers interact with artifacts reposi-

tory, which should include the components and de-

vices for the generation of the SPL products. To build

SPL products, we can also access distributed reposi-

tories.

The sequence alignment process can perform the

complete alignment of sequences (called global align-

ment) or only compare fragments of the sequences

(called local alignment), as can be seen in Figure 3(a).

The sequence alignment can also indicates the degree

of similarity between two sequences (called pairwise

alignment) or more than two sequences (called mul-

ConnectingFeatureModelsandOntologiesinSoftwareProductLines

161

Figure 2: Product line Architecture.

Figure 3: Global, local, pairwise and multiple aligning.

tiple alignments). Both local and global alignments

can be performed with two or more sequences, as can

be seen in Figures 3(b) e 3(c).

This paper focuses on ’pairwise local aligning’,

shown in Figure 3(c). The ’pairwise local aligning’

can be executed by the family of algorithms known as

BLAST (Basic Local Alignment Search Tool). This

family can be divided into ﬁve different programs, as

can be seen in Figure 4.

Figure 4: BLAST family of programs.

In our example, we aimed to deﬁne which of

BLAST is more suitable for a speciﬁc type of ’pair-

wise local aligning’. To address this issue, we iden-

tiﬁed the main required parameters to perform each

BLAST algorithm. These required parameters were

represented as restrictions in our ontology. So, the

three following parameters were considered: (i) the

input sequence type, (ii) the format of the compared

sequence and (iii) the database used for the query.

Using only a feature model, we cannot express the

semantic about our domain as we can do using on-

tologies. To solve this lack of semantics of our fea-

ture model, we use a Sequence Alignment Ontology.

This ontology was adapted from the myGrid ontol-

ogy (Wolstencroft et al., 2007). We use only part of it

and add some extensions in order to make the ontol-

ogy usable in conjunction with our feature model. It

is important to emphasize that we have some work

on this topic using other ontologies, mainly CeLO

(Matos et al., 2010), (Silva et al., 2012), but in this

example, the myGrid Ontology is more suitable.

In our approach, the Variability Manager is repon-

sible for the connection between the models. This

manager works according to the following steps:

1. It takes the mandatory features from the feature

model and presents them to the user. So, the user can

make the choice of a desired subfeature of the manda-

tory features.

2. The manager picks the features that were se-

lected by the user and accesses a simple XML ﬁle.

This ﬁle has a mapping between the feature model

and the classes of the ontology. Currently, this XML

ﬁle is created manually, with the help of a domain

specialist. However, we are planning to use some on-

tology alignment techniques, tailored in order to be

used with a feature model.

3. With the corresponding class name, the Vari-

ability Manager can access this class in the ontology

and can get its restrictions.

4. As the restrictions refers to features that should

be selected, once they are related to those features that

were selected at an earlier time, the variability Man-

ager consults the XML mapping and gets the feature

name of the restriction.

5. Finally, the new set of features to be selected is

displayed to the user, in accordance with the restric-

tions which were found.

These ﬁve steps should be performed until the

conﬁguration of the features necessary for a given

product is completed and there are not more restric-

tions to be found.

It is important to notice that, in this process, the

inference mechanism in the ontology can act in or-

der to improve the selections. The use of inferences

in ontologies can be considered one of the main ad-

vantages of this modeling technique domain and may

be used in our approach to ﬁnd other links between

the different features available, in order to provide a

better selection of the features in our SPL.

In our example, the mandatory features and its

ICEIS2012-14thInternationalConferenceonEnterpriseInformationSystems

162

Figure 5: Feature selection.

Figure 6: Restriction of ’pairwise local aligning’.

subfeatures are shown to the user using selection

boxes. As illustrated in Figure 5, the subfea-

ture ’pairwise local aligning’ was selected. After

that, the Variability Manger uses the XML mapping

ﬁle to obtain the class name of the feature ’pair-

wise local aligning’. It consults the restrictions of the

class in the ontology. This class has two restrictions,

as can be seen in Figure 6.

Using these restrictions, the Variability Man-

ager consults the XML to transform the classes

of the restrictions into features. Next, in

our example, is shown to the user a selec-

tion box with the subfeatures of ’bioinformat-

ics algorithm’ and an information to select the feature

’BLAST Basic Local Alignment Search Tool’.

When the BLAST bioinformatics algorithm is se-

lected, the Variability Manager repeats the step 2, as

was proposed before, and obtains the restrictions of

the BLAST. According to the BLAST restrictions, the

user must select: (i) the format of the input sequence,

(ii) the format of the sequence to be compared with

and (iii) the format of the sequence database. Fi-

nally, after the selection of this features, the Variabil-

ity Manager has to associate all the selected features

and its restrictions. After that, it needs to ﬁnd, in the

ontology, the most appropriate class that has these re-

strictions.

For example, if the user selects: (i) a protein input

sequence, (ii) a protein sequence format to compare

with and (iii) a database with sequences of protein, the

Variability Manager can inform, through the ontology

analysis, that BLASTP is the most compatible class

with the previously selected features.

It should be emphasized that this work is only the

ﬁrst step towards the proposed SPL. Our research en-

compasses all the Sequence Alignment SPL.

5 CONCLUSIONS

Throught this paper, we present an approach to con-

nect a feature model and ontology using a Variabil-

ity Manager and a XML mapping ﬁle. It aims to

obtain the advantages of both domain modeling tech-

niques. To illustrate our approach we applied this ap-

proach in the bioinformatics context, attacking the se-

quence aligning problem. Through our example, we

illustrated how an ontology can improve the feature

model, providing additional and relevant information

for the domain of the proposed SPL.

For further work, the issues below remain to be

resolved in this research: (i) improve and evolve our

models, incorporating into this all the tasks needed

in a Sequence Alignment SPL; (ii) generate a semi-

automatic matching between the feature model and

the ontology and (iii) improve the application inter-

face by which the user must select the features.

REFERENCES

Asikainen, T., M

annist

o, T., and Soininen, T. (2007). Kum-

bang: A domain ontology for modelling variability in

software product families. Adv. Eng. Inform., 21.

Czarnecki, K., Peter Kim, C. H., and Kalleberg, K. T.

(2006). Feature models are views on ontologies. In

Proceedings of the 10th International on Software

Product Line Conference, pages 41–51, Washington,

DC, USA. IEEE Computer Society.

Gruber, T. R. (1993). Toward principles for the design of

ontologies used for knowledge sharing. In Formal On-

tology in Conceptual Analysis and Knowledge Repre-

sentation. Kluwer Academic Publishers.

Johansen, M., Fleurey, F., Acher, M., Collet, P., and Lahire,

P. (2010). Exploring the synergies between feature

models and ontologies. In International Workshop on

Model-driven Approaches in Software Product Line

Engineering (MAPLE 2010)(SPLC’10 (Volume 2)),

volume 2 of SPLC’10 (Volume 2), pages 163–171,

Jeju Island, South Korea. Lancester University.

Matos, E. E., Campos, F., Braga, R., and Palazzi, D. (2010).

Celows: An ontology based framework for the pro-

vision of semantic web services related to biological

models. J. of Biomedical Informatics, 43:125–136.

Mendes, L. F., Silva, L., Matos, E., Braga, R., and Campos,

F. (2011). Sasagent: An agent based architecture for

search, retrieval and composition of scientiﬁc models.

Comput. Biol. Med., 41:449–462.

Silva, L. M., Braga, R., and Campos, F. (2012). Composer-

science: A semantic service based framework for

workﬂow composition in e-science projects. Inf. Sci.,

186:186–208.

Wolstencroft, K., Alper, P., Hull, D., Wroe, C., Lord, P. W.,

Stevens, R. D., and Goble, C. A. (2007). The my-

grid ontology: bioinformatics service discovery. Int.

J. Bioinformatics Res. Appl., 3:303–325.

ConnectingFeatureModelsandOntologiesinSoftwareProductLines

163