This project aims to build a MetaIntegrator for integrating databases on the Internet. To realize large-scale integration over this "deep Web," the research pursues the "shallow integration" philosophy, which does not rely on deep semantic annotations or deep probing of sources. In particular, it takes a holistic approach, which essentially leverages the challenge of large-scale itself as an opportunity, by resorting to hidden statistical regularities for enabling shallow integration. Specifically, this project focuses on three tasks: (1) Deep Web survey: To characterize the nature of the deep Web (as pertinent to semantic integration), this project extensively surveys databases on the Web. (2) Source selection: Since there are so many potential sources, "what sources should be involved in answering a given query?" This project thus develops the dynamic selection of sources for matching a query. (3) Schema integration: Since sources are heterogeneous, in querying the problem "how does an attribute in one source correspond to that in another?" Needs to be resolved. This project thus studies the matching and unification of schemas across myriad sources. Information on this project will be posted on the Web (http://eagle.cs.uiuc.edu/metaquerier) and the results are expected to have broad impact by advancing the understanding of dynamic, large-scale integration and develop a set of general techniques, and constructing a MetaIntegrator for providing access to (selected domains of) the deep Web. These fundamental results are likely to have technology transfer into a broad spectrum of Web applications.
