Data Integration Based on Data Conversion and Restructuring

Due to the development of the World Wide Web, the integration of heterogeneous data sources has become a major concern. Appropriate architectures and query languages have been proposed but the problem of data conversion remains largely unexplored. We present the YAT system for data conversion which provides tools for the speciication and the implementation of data conversions among heterogeneous data sources. It relies on a middleware model, a declarative language to described conversion/integration programs, a graphical representation for the language, and several mechanisms allowing to easily reuse existing programs. The model is based on named trees with ordered and labeled nodes. Like semistructured data models, it is simple enough to facilitate the representation of any data. The main originality of the model is its ability to capture various levels of representation. A YAT model can be instantiated into another, more speciic (eventually "ground"), model. This novel feature is essential rst for allowing customization of conversion/integration programs. Also, it is a key component of the type veriication that can be used to validate conversion programs. Finally, it is central to allow conversion programs to be combined (in parallel) or composed (sequentially) in a coherent manner. The YAT conversion language (YATL) is declarative, rule-based and features enhanced pattern matching facilities and powerful restructuring primitives. It allows to preserve or reconstruct the order of collections. We also present the architecture, implementation and practical use of the YAT prototype.