Digital Libraries (DLs) are extremely complex information systems that integrate ¯ndings from disciplines such as hypertext, information retrieval, multimedia services, database management, and human-computer interaction. Designers of DLs are often multidisciplinary teams, which include library technical sta® and computer scientists. Wasted e®ort and poor inter-operability can therefore ensue, raising the costs of DLs and risking the °uidity of information assets. To alleviate these problems, we use a model-driven approach for the design and automatic generation of code for DLs. In particular, we use a Domain Speci¯c Visual Language (DSVL) made of four diagram types (collection, structural, service and societal) which describe the di®erent aspects of a DL. We have built a code generator able to produce XUL code from the design models for the DL user interface. This XUL code integrates prede¯ned components for the di®erent services, according to the model speci¯cation.
¤Work supported in part by the EC's Human Potential Programme under contract HPRN-CT-2002-00275, SegraVis, and the Spanish Ministry of Science and Education, projects MD2 (TIC200303654) and MOSAIC (TSI2005-08225-C0706). environments, and sets of services objecti¯ed to meet users' needs. Underlying all these de¯nitions, there is the agreement that DLs are fundamentally complex, due to their interdisciplinary nature. They are usually built from scratch using specialized architectures that do not bene¯t from past design experiences. In addition, formal models, which support research and development in most computer science sub¯elds, are surprisingly unaccounted for within the DL literature. This lack of formality leads to branching e®orts and has made interoperability one of the most crucial problems faced by the DL ¯eld.
In Model-Driven Development (MDD), models are the
primary assets, from which code is generated for a particular
platform. Moreover, many User Interface Description
Languages (e.g. UsiXML, XAML and XUL) have been
introduced so far that address di®erent aspects of a User Interface
(UI) [
Our approach has the following advantages. The use of a language that is visual makes it easier for people to interpret the model, while being domain-speci¯c leaves less room for misunderstandings. The language has been speci¯ed by metamodeling, what makes it possible to build tools providing automatic built-in syntactic and semantic checks that guide the DL designer in ¯nding inconsistencies and gaps in the designs, and making sure that a change in a model is automatically re°ected in other parts where it is relevant. In addition, although visual, the language is also formal, thus it is possible to simulate and analyze the models to demonstrate, understand and verify its behavior. Finally, deriving code from the models minimizes the number of errors in the ¯nal DL implementation.
The paper is organized as follows. Section 2 presents some related research, while in section 3 we sketch the overall proposed architecture. Section 4 presents the VisMODLE language, for which a modeling environment has been generated, as explained in section 5. In section 6 we show the generation of the XUL code for the DL's UI. Finally, section 7 ends with the conclusions and future work. Throughout the paper we use a toy example of a university library.
Formal models for DLs are rarely found. Wang [
Lee et al. [
Our work investigates and empirically evaluates a new MDD framework that allows DL designers to model ideas and mechanisms speci¯c to interaction in the DL domain, and to transform them to the ¯nal (compilable) source code.
To improve acceptability and interoperability, our framework makes °exible use of existing standard speci¯cation sub-languages for representing DL concepts. Therefore, most of the model primitives are de¯ned as XML-based elements, which can enclose other sub-languages that help to de¯ne DL concepts. In more detail, MIME1 types constitute the 1Multipurpose Internet Mail Extensions basis for encoding elements (e.g. documents) of a collection. The XML User Interface Language (XUL) is used to represent the DL visual interface. It provides a simple and portable de¯nition of common widgets, thus drastically reducing the software development e®ort for visual interfaces. However, the framework is general and other UI description languages can be used.
In this section we present VisMODLE, a DSVL oriented to the design of DLs, as well as a running example, called Library, to show the overall process starting from the basic entities of the model. The example is kept small for presentation purposes, although our system has been able to deal with real-life DLs.
In VisMODLE the speci¯cation of a DL encompasses four complementary dimensions or viewpoints, including: multimedia information supported by the DL (Collection Model); how that information is structured and organized (Structural Model); the overall behavior of the DL (Service Model) and the di®erent societies of actors and groups of services that act together to carry out the DL behavior (Societal Model). The complete metamodel is shown in Figure 2.
Collections are sets of elements. They can model both static (e.g. text) and dynamic content (e.g. presentation of a video). Our running example includes a collection model (not shown for space constraints) that de¯nes a collection (called Library) of two documents: long1.pdf and long2.pdf. The Structure speci¯es the way in which parts of a whole are arranged or organized. In DLs, structures can represent hypertexts, taxonomies, system connections, user relationships, and containment. The window to the right in Figure 5 shows the structural model for the running example. Thus, the collection Library is made of documents structured with Publication, Author and Title metadata information (i.e. three Struct elements). Metadata entities are linked together with the node relation (organized as a tree) and linked to a collection by a metadata link type relation. Services as scenarios tell what happens to the elements in the collection and through the structures. Taken together the services describe scenarios, activities, tasks, and operations, and those ultimately specify the functionalities of a DL. Human information needs, and the processes of satisfying them in the context of DLs, are well suited to description with services, including these key types: fact-¯nding, learning, gathering, and exploring. In our example we make available two basic services: Front Desk and Search. The Front Desk is responsible for managing communications between actors and it is asynchronous (sync attribute is set to nowait ), while the Search service executes queries on the DL and it is synchronous.
Finally, a Society is a set of entities and the relations between them. The entities include actors as well as hardware and software components, which either use or support services. A society is the highest-level component of a DL, which exists to serve the information needs of its entities and to describe the context of its use. DLs are used for collecting, preserving, and sharing information artifacts between society members. Figure 3 shows the societal model for our example. It involves two Actors: Student and Librarian. The scenario represents a Student trying to borrow a paper from the Library; he interacts with the Front Desk service requesting the paper and obtaining a response message about its availability. The Front Desk service forward the borrow request to the Librarian actor. Then, the Librarian sends a doc request message to the Search service, which queries the document collection (get operation) using metadata information provided by the borrow request, and waits the result to send back the response. The service returns an is available boolean message which is propagated as a response to the Librarian and eventually to the Student.
We have generated a modeling environment for VisMODLE
using the metamodeling tool AToM3 [
The model of the speci¯c DL drives the UI generation. The UI layout is generated mainly from the information provided by the structural and collection models. In particular, code is automatically generated by \LibGen" when the DL designer modi¯es these types of models, thus immediately seeing the result of a change. It is also possible to produce code in a batch style. The UI events are managed by invoking the appropriate services according to the imported XUL templates, as shown in Figure 1.
The generated UI is built upon a set of XUL template ¯les that are automatically specialized depending on the attributes and relationships designed in the modeling phase. The layout template for the UI is divided into two columns (see Figure 6, that shows the UI generated for the actor Librarian in our example). The left part is made of three boxes: Collection, MetaData and MetaData Operations. The right part manages visualization and multimedia information obtained from documents. The basic features provided with the UI templates are: document loading and visualization, and metadata organization and management.
The Collection box (\B" in the ¯gure) manages the visualization of documents. The list of documents is obtained from the attribute documents of the collections speci¯ed in the DL design (i.e. long1.pdf and long2.pdf in the collection Library of our example). The visualization template works according to the (MIME) data type speci¯ed in the modeling phase. In fact, by selecting a document, the corresponding ¯le is uploaded and visualized (starting the appropriate viewer, see \A" in the ¯gure) within the generated UI, and in addition can be afterwards managed (print, save. . . ). The MetaData box (\C" in the ¯gure) manages the tree structure used for representing the metadata information. The tree structure of the metadata is generated according to the metadata categorization modeled by the designer in the structural model (model to the right of Figure 5). In our example a tree has been generated with a root metadata node Publication that contains two children metadata nodes Title and Author. The XUL template contains all the basic layout and action features for managing a tree structure. The MetaData operations box (\D" in the ¯gure) is activated by clicking on a node of the metadata box. It manages metadata operations, such as insertion, deletion or editing. These automatically generated interfaces have a standard template design that can be integrated by the users/designers simply adding code or additional style templates in the appropriate template placeholders. In VisMODLE, the societal model drives the linking among the visual interface templates (XUL code) and the services implementation (Java template code for browsing, indexing, searching, ...) in order to generate the full DL system (actors, services, collections and their interactions).
In this paper we have presented a novel MDD approach for the generation of DLs. For this purpose, we have designed the VisMODLE DSVL, generated a customized modeling environment, and built a code generator able to produce the full application. In the present paper, we have focused on the automatic generation of the UI, using XUL. Currently, we are working in extending VisMODLE with a behavioral diagram, and in adding analysis and simulation capabilities to the framework. Moreover, we intend to support XDoclet for the speci¯cation of the VisMODLE services. It allows automatic code generation, compliant with a standard de¯nition which simplify coding for various technologies, such as: Java, Web Services and Web Portals.