eXtensible Relational Databases: a relational approach to interoperability

Successful data integration relies on knowledge of (i) the structural shape of constituent data sources, and (ii) semantic knowledge of how structural terms map together, or how structure maps to structure. In this paper, we propose a simple extension to the relational model which allows these types of metaknowledge to be stored directly within relational sources and then incorporated into queries over these sources. This extension is termed the eXtensible Relational Database (XRDB) due both to affinities of the concept of metaknowledge with semantic web ideology, as well as due to use of the framework in storing active XML metadata – semantic metadata with functionality – in relational databases. While XRDB entails storing data from two fundamentally different conceptual levels within the relational model, the query language we propose is a simple extension of ordinary SQL and allows for seamless usage of both data and metaknowledge. A crucial advantage of the extended language is that it allows dynamically generated mediated views to be computed from metaknowledge stored in the data sources. These views address problems ranging from semantic heterogeneity to structural dissimilarity, and even allow us to quantify dissimilarity of data sources. A prototype implementation of our framework exists, and we briefly describe its underlying architecture. We conclude the paper by discussing how our query language compares to existing metadata integration languages.