Model driven design and data integration in semantic web information systems

The Web is quickly evolving in many ways. It has evolved from a Web of documents into a Web of applications in which a growing number of designers offer new and interactive Web applications with people all over the world. However, application design and implementation remain complex, error-prone and laborious. In parallel there is also an evolution from a Web of documents into a Web of `knowledge' as a growing number of data owners are sharing their data sources with a growing audience. This brings the potential new applications for these data sources, including scenarios in which these datasets are reused and integrated with other existing and new data sources. However, the heterogeneity of these data sources in syntax, semantics and structure represents a great challenge for application designers. The Semantic Web is a collection of standards and technologies that offer solutions for at least the syntactic and some structural issues. If offers semantic freedom and flexibility, but this leaves the issue of semantic interoperability. In this thesis we present Hera-S, an evolution of the Model Driven Web Engineering (MDWE) method Hera. MDWEs allow designers to create data centric applications using models instead of programming. Hera-S especially targets Semantic Web sources and provides a flexible method for designing personalized adaptive Web applications. Hera-S defines several models that together define the target Web application. Moreover we implemented a framework called Hydragen, which is able to execute the Hera-S models to run the desired Web application. Hera-S' core is the Application Model (AM) in which the main logic of the application is defined, i.e. defining the groups of data elements that form logical units or subunits, the personalization conditions, and the relationships between the units. Hera-S also uses a so-called Domain Model (DM) that describes the content and its structure. However, this DM is not Hera-S specific, but instead allows any Semantic Web source representation as its DM, as long as its content can be queried by the standardized Semantic Web query language SPARQL. The same holds for the User Model (UM). The UM can be used for personalization conditions, but also as a source of user-related content if necessary. In fact, the difference between DM and UM is conceptual as their implementation within Hydragen is the same. Hera-S also defines a presentation model (PM) which defines presentation details of elements like order and style. In order to help designers with building their Web applications we have introduced a toolset, Hera Studio, which allows to build the different models graphically. Hera Studio also provides some additional functionality like model checking and deployment of the models in Hydragen. Both Hera-S and its implementation Hydragen are designed to be flexible regarding the user of models. In order to achieve this Hydragen is a stateless engine that queries for relevant information from the models at every page request. This allows the models and data to be changed in the datastore during runtime. We show that one way to exploit this flexibility is by applying aspect-orientation to the AM. Aspect-orientation allows us to dynamically inject functionality that pervades the entire application. Another way to exploit Hera-S' flexibility is in reusing specialized components, e.g. for presentation generation. We present a configuration of Hydragen in which we replace our native presentation generation functionality by the AMACONT engine. AMACONT provides more extensive multi-level presentation generation and adaptation capabilities as well aspect-orientation and a form of semantic based adaptation. Hera-S was designed to allow the (re-)use of any (Semantic) Web datasource. It even opens up the possibility for data integration at the back end, by using an extendible storage layer in our database of choice Sesame. However, even though theoretically possible it still leaves much of the actual data integration issue. As this is a recurring issue in many domains, a broader challenge than for Hera-S design only, we decided to look at this issue in isolation. We present a framework called Relco which provides a language to express data transformation operations as well as a collection of techniques that can be used to (semi-)automatically find relationships between concepts in different ontologies. This is done with a combination of syntactic, semantic and collaboration techniques, which together provide strong clues for which concepts are most likely related. In order to prove the applicability of Relco we explore five application scenarios in different domains for which data integration is a central aspect. This includes a cultural heritage portal, Explorer, for which data from several datasources was integrated and was made available by a mapview, a timeline and a graph view. Explorer also allows users to provide metadata for objects via a tagging mechanism. Another application is SenSee: an electronic TV-guide and recommender. TV-guide data was integrated and enriched with semantically structured data from several sources. Recommendations are computed by exploiting the underlying semantic structure. ViTa was a project in which several techniques for tagging and searching educational videos were evaluated. This includes scenarios in which user tags are related with an ontology, or other tags, using the Relco framework. The MobiLife project targeted the facilitation of a new generation of mobile applications that would use context-based personalization. This can be done using a context-based user profiling platform that can also be used for user model data exchange between mobile applications using technologies like Relco. The final application scenario that is shown is from the GRAPPLE project which targeted the integration of adaptive technology into current learning management systems. A large part of this integration is achieved by using a user modeling component framework in which any application can store user model information, but which can also be used for the exchange of user model data