Companies focused in the design and implementation of Web based systems, which are also interested in the adoption of state-of-theart Model-Driven Web Engineering and Software Product Line Engineering techniques, are nding di culties adapting the requirements of their domains to the already available tooling. This article explains how to successfully combine these two approaches in an industrial context. The techniques described in the case-study show how model-driven product line engineering can be used to provide domain knowledge encapsulation, and automatic generation of the products. These products include both, Content-Manager infrastructures to populate the system's contents, and automatically adapted Web frontend prototype generation for di erent devices and user pro les.
The key idea of Software Product Lines (SPL) is the use of a common set of
assets for product derivation [
Industrial companies focused on the development of Web Systems and content providers for Web Systems are trying to improve their daily activities looking for new ways to perform their processes. Apart from a considerable lack of strategic reuse of their Web Systems components, current state-of-the-practice in many of the companies of this market-segment experiment two main problems: On the one hand, the non-existence of a uni ed way to introduce the contents derives in an unnecessary waste of time for the employees to learn new technologies and feel comfortable with the new platforms. On the other hand, a rapid prototyping platform is also desirable for showing their customers a working prototype at an early stage.
This article presents a solution for these two problems, achieving at the same time other important goals such as company knowledge encapsulation, automatic and strategic reuse of their components and the bene ts of an architecture-driven infrastructure for the company that follows a Model-Driven approach.
Conmunica Mediatrader is a company with expertise in contents, press clipping and product placement. Located in Spain, it belongs to a multimedia leading group in the country. They are specialized in the development of customized applications for their clients. The work described in this paper allows Conmunica Mediatrader to minimize the e orts and costs these developments imply, making their daily work easier.
This paper is structured as follows: Section 2 makes an introduction to the PLUM tool suite that is necessary to understand SPL main assets. Section 3 describes the proposed process to be followed by the company using this approach and sections 4 and 5 describe in detail the implemented models and how the domain variability is interpreted for product generation.
PLUM is an open-source tool suite for the design, implementation and management of Product Lines that follows a Model-Driven Software Development approach. PLUM is a domain agnostic tool suite. Its aim is to ll all the needs of di erent vertical domains with regard to SPL Engineering.
Using PLUM, domain's products variability is captured in what is called
a Decision Model. Decision Modeling implies analyzing domain variability in
terms of decisions and establishing dependencies among them. This technique,
which comes from previous European research projects such as FAMILIES[
The Decision Model is input for the Variability Resolution Process at which concrete Application models are derived. The Variability Resolution Process consists in giving values to the Decisions in order to de ne the desired product. In this way Application Models represents concrete products of the product line. This concept is also illustrated in Figures 3 4 and 5 regarding to speci c Decision models and Application models of the present case-study.
The transformation of these Application models into the desired output (source code, documentation or other models) is possible through what are called the Flexible Components. These reusable assets interprets domain variability in order to generate the required products. Another important feature of PLUM consists in Work ow de nition. The Work ows establish the process for product generation and deployment. The Flexible components are called from the workows for code generation, and other deployment processes such as generated code compilation, retrieving dynamic data, launching external processes etc. could be started if needed.
PLUM is integrated in Eclipse[
Adoption of SPL architecture and SPL Engineering practices demands a strong company commitment. These initial e orts have to be taken into account in the return on investment (ROI) measurement. The economic aspect is out of the scope of the present article, nevertheless, it is important to notice that company way of working is going to be changed slightly. The company, Conmunica Mediatrader, is intended to adopt an architecture-centric approach that uses a Model-Driven Software Development paradigm. Our proposed Web system lifecycle is separated in two phases, a Generation phase and a Management phase, as shown in Figure 1 and gure 2 respectively. Following subsections aim to explain these implemented phases paying special attention to the rst one as is more relevant in the present article context.
First product generation sub-phase consists in the requirements gathering with the customer. The main stakeholders reach an agreement with the customer about three main aspects: Web system structure, usability and accesibility requirements and the devices that should be supported. With this information, Web, potential users and device Decision models could be used to derive instances of concrete Application models for each of them. The content architect starts the variability resolution process of the Web model obtaining an instance of the desired Web structure. The usability expert, on their part, makes use of the User application models library looking for user pro les that ts customers expectations of potential users. If a new user pro le is needed, he starts the variability resolution process of the User Decision model. Finally, the technical department check if the devices the Web should support are in the Devices Application models library and, also, they start a variability resolution process of the Device Decision model if not found.
PLUM tool suite allows an assisted and guided variability resolution process of the Decision models. In addition, the separation of the di erent roles allows that the three di erent Decision models are de ned using the domain concepts that each expert understand. For example, the usability expert will decide about potential users possible cognitive impairments while the technical department will decide about connections' quality of service parameters.
As output of the Requirements gathering phase, apart from the application models, the requirements themselves are also inputs for the SPL maintenance team. They are responsible of detecting needs of SPL assets further development. Some requirements could demand supporting new unexpected variability in the decision models, and that could have a direct relation to the exible components that interprets this variability for Web prototype, database or content manager automatic generation. In summary, requirements interpretation a ects SPL assets evolution, in this way the company enrich their reuse capability at the same time that they ll customer needs.
The following phase, the Product Line Work ow, is an automatic process initialized by the PL Execution engineer. Given the Web model de ned by the content architect with the customer, a database for storing the Web contents is automatically derived. This is achieved by model-to-text transformations, implemented in the Flexible Components, which takes the Web model and generate a database schema with only the needed tables and columns. The execution of a dened PLUM Work ow allows not only to generate the code of the database, but also to deploy this code. Using o -the-self Work ow components, the database schema is translated to SQL queries and directly executed in a selected server. In the same manner, the Web Content Manager used for accessing the database is generated. The generated Content Manager is speci c for the Web systems domain so its aim is to o er an usable way to introduce Web contents.
Web style, source code language, and how to present the contents vary in terms of user pro les and target devices. The automatic process of variability interpretation through the Product Line assets allows the Adapted Web prototypes generation. The de ned Web characteristics are represented in the automatically generated nal output obtaining a working Web prototype that could be even shown at very early validation phases.
Once the products are generated at the previous phase, the Content Manager can be immediately used for contents introduction. However, the adapted Web prototype should go through new de ned phases before being in production. The rst phase is the validation phase at which the customer or the testers re ne Web systems requirements. Their feedback could be very valuable for possible changes in the previously taken decisions.
On the other hand, the generated prototype could be used by the developers to start a incremental prototyping methodology for the development of the Web system. The company deliver to the customer a working Web that could be put online when they consider that this should be done. In order to close the Web system life-cycle, nal users could give feedback to the customer that could derive in small code changes or in requirements rede nition.
In order to model the variability of an environment with user accessibility and multi-device support requirements, we have identi ed the need of three Decision models to capture the variability of a Web system. The Web Decision model represents the structure and contained information of a Web site. The User Decision model de nes user pro les in terms of accessibility aspects. Finally, the Device Decision model represents the main characteristics of the devices that could be used to display the Web system.
The de nition of separated models allow us not only to separate the concepts in areas, but also separate its functionality in order to generate tailored solutions. In other words, this makes it possible to create di erent Web sites adapted for di erent devices or users maintaining the same Web Application model. It is only needed to change the selected Devices and Users Application models as inputs for executing the adapted Web generation Work ow.
Following subsections explain the implemented Decision models in detail. Due to space restrictions only the most signi cant decisions of these Decision models are expanded at gures 3, 4 and 5.
The Web Decision model has been de ned in order to represent the general structure of a Web site. It can be seen as the so-called site map that every single Web site in the Internet has or should have in order to represent its content. The rst step when de ning a Web, resolving the model variability in a certain Application Model, is to set up a "Name" and a "Description" with the aim of distinguishing the site from others. It is also essential to select the type of multimedia les the site hosts (in case it does), using the choice decision "Multimedia", which possible values are: Audio, Photos, Video and Advertising.
Another important and valuable feature these days is the capability of
displaying the content of a Web in several languages. The Web model de ned
considers this possibility within the group of decisions named "Locations",
following the rules for Internationalization and localization [
The biggest group called "Structure" collects the list of "Categories" that will constitute the site, hence de ning the schema or the site map. For each category it is mandatory to de ne a "Name", the number and type of the sections that it will contain and lastly assigning a "Name" for each section included. Section prede ned in the model for this purpose are: News, Events, Contacts, Interviews, Multimedia Galleries, Links, Advertising and Timetable. For each type of section it is also possible to choose between a list of features that could be available or not in the nal site. For instance, the section "Event" o ers the possibility of including associated locations for an event (which take place in a certain date), associated events, a reference to the source of information or a link to buy tickets online. Besides, each section includes information by default which is always present when de ning this type of section, independently of the Web site. Taking this into account, an "Event" section always has a title, a description (divided into paragraphs) and, in case they were selected previously, multimedia les.
The presented solution for Modeling Web Systems does not diverge from the
classical Human Computer Interaction engineering approach. The Web System
architecture is modeled in terms of a hierarchy of categories and these categories
have di erent sections that vary in its nature. This hierarchy can be seen in other
approaches as UIML [
Another important issue to be considered in web-modeling is User Task
Modeling [
User modeling is a step that is explicitly declared in many user-centered design
processes[
The User Model presented in the current article de nes a user pro le based on the accessibility requirements (in terms of age, experience and disability) that the nal user could have at the moment of accessing to a certain Web site. The implementation of this model has been done taking into account one of the requirements de ned by Conmunica Mediatrader that consists in the fact that the generated Web sites have to adapt their contents depending on the characteristics or disabilities of the nal user. As the Figure 4 depicts, the main features that constitute a user pro le are "Age", "Impairment" and "Skill level". These features are considered relevant as they are the main characteristics of a person that could have a strong in uence in the way the results are generated or displayed.
The rst decision to take in our model, is to select a range for the "Age" feature. Once the range is selected for the Age decision, a set of default values are automatically assigned to the rest of decisions through the de ned domain dependencies. For instance, if the value selected is "Elder", it could be taken for granted that the user may have visual impediments, hearing problems, and some kind of cognitive and motor impairment.
The following decision, and probably the most important one, is related with physical impairment or disabilities. The disabilities have been organized in ve general groups: visual impairment, cognitive impairment, hearing impairment, motor impairment and language impairment. The values selected in these groups are the ones with most impact on the nal Web to be generated. For example , if "low vision", "di culty using mouse" and "deafness" are selected, then the content in the Web will be generated with big font sizes in both texts and buttons, without scroll bars and multimedia les, such as videos or recordings, will be subtitles or omitted. Not only the values selected directly are taken into account but also are the values implicit within each group. Finally, the last decision to make is to choose the skill level of the user. Depending on the level of knowledge of the user, the type of content to be created varies in a great manner. Obviously, a Web Site with speci c thematic in which the user has well de ned peculiarities di ers completely from a general purpose Web Site.
This User Model is based on the Accessibility ontology[
The development teams of Conmmunica are used to develop Web Systems for
di erent device families. This fact has to be considered for the automatic
generation of the adapted Web, therefore, a Device Model capable of storing all
the device properties is needed. The devices that will be implemented with this
model are mobile devices such as mobile phones or pda's, and Digital Terrestial
Television (DTT). Approaches regarding device modeling aimed to User
Interfaces adaptation can be found at WURFL [
The Flexible Components are the assets that interpret product variability for speci c product generation. In this section, in order to illustrate the products generation we present an example. The example is based on a theatre: the "Stacey Theatre".
The content architect de nes the Web structure through the variability resolution of the Web Decision Model. The decision making process with the Stacey Theatre owner's requirements allows to obtain a Web application model that denes the desired Web system: The Stacey Theatre Web must have the possibility of being visualized in three languages: English, Spanish and French. Categories and the sections are also de ned in the following manner: events and interviews, multimedia galleries of photos and videos, the history of the theatre, contact information, useful links, etc.
The products generated with our solution are a Web site and the corresponding Content Manager application to manage the information contained in the mentioned Web.
The Content manager Web application facilitates the labour of updating the contents of the Web site itself. The automatic generation of this Web system related product represents a competitive advantage for Conmunica Mediatrader. Figure 6 depicts the Content Manager application that is automatically generated from a certain Application Model, more precisely, it shows the Content Manager for the "Stacey Theatre" example. The name displayed at the top is automatically obtained from the name established in the corresponding application model.
The menu on the left side has been ful lled also automatically from the values available in the "Stacey Theatre" application model, more concisely, from the names of the categories and sections de ned within them. Clicking on each item, the application will show the elements that must be ful lled for the current section and which are totally dependant on the features previously selected for each section in the application model. The application model for the current example de nes that the Event section (Plays in the left menu) will not include the "Associated locations", "Associated Events" and "References to sources" elements mentioned in the Web decision model, but it will include a link to buy online tickets. The rest of the elements which are shown in the gure, such as title or description, are the elements considered as mandatory in every single Web site to de ne an "Event" element, in other words, the minimum information necessary to de ne the event. For every section de ned in the model, it is considered that its description is divided into paragraphs, which can be reorganized whenever necessary (with aid of the actions buttons). This implies that is possible to de ne two kind of paragraphs: plain text and multimedia le (in case the site hosts these les). Finally, all the text data is required in di erent languages as the links at the top of the main area inform. The languages displayed in these tags depend on the "Locations" de ned in the application model, being English, Spanish and French for the "Stacey Theatre" example.
The adapted Web obtained from the SPL architecture is not intended to be a
Web system ready for production. Despite of that, the obtained Web is a working
prototype that allows an instant visualization of any under-design Web system
that is being modeled. Apart from that, the generated prototype is adapted
using information from speci c user and devices models obtaining by this mean
a valid approximation of the nal result. In adaptable Web Systems research
literature, user context and device constraints are taken into account to change
not only the Web System but also Service compositions. Using a Product Line
approach for achieving these goals is studied at [
Figure 7 depicts the generated Web for the "Stacey Theatre" example, where it can be appreciated how di erent the results are depending on the targeted user and device. Speci cally, it shows how the Web could change their contents presentation for elderly people, mobile phones and DTT (Digital Terrestial Television).
With regard to the user adaptation, the most common case of a generated Web is the one in which the output is targeted to someone who does not need a special adaptation. Otherwise, when thinking about adapted Web site, not only we have to focus on people with physical or cognitive disability, but also we have to realize that there are some factors like the age, which determine the way of visualization of certain contents. Because of this, if the generation of a web site for elderly people is required, it is important to consider that they can have some kind of vision impairments or motor impediment, and consequently, it is necessary to provide them with the information adapted to their needs.
Many types of devices can be found nowadays in the market, ranging from devices limited to voice processing to devices with great multimedia processing capabilities. This Device variability is also an important aspect for the Web system adaptation. Supported technologies, multimedia processing capabilities and input/output components determines how the contents are displayed.
The current paper presents how to apply Model-Driven Product Line engineering using Decision Modeling techniques in order to capture Web systems domain variability. Our approach accounts for user pro les and devices as they directly a ect how concrete Webs are displayed.
We have presented the rst results of the Industrial usage of the mentioned approach. The company, Conmunica Mediatrader, bene ts from Content Manager infrastructure and adapted Web frontend prototype automatic generation. These are the Web systems related products obtained from the Product Line implemented with the PLUM tool suite in this case-study.
The work reported here has been partially sponsored by the SPRI (Association for the Promotion of Industry) within the Adapta Project (INTEK-BERRI R&D programs for supporting Basque Country companies) and the "Ministerio de Industria, Turismo y Comercio" within the FLEXI FIT-340005-2007-37 (ITEA2 6022) European project.