Abstract
Against the backdrop of the rapid proliferation of heterogeneous databases, the challenge of data integration lies in enabling multiple databases to interoperate, thereby providing a single point of entry for applications. To achieve this, the data in the sources must first be mapped to the integration schema that is to be exposed to applications. “Mappings,” or logical correspondence rules, allow this to be done formally; current research aims to automate their generation. This need is particularly significant in “data lakes,” where thousands or tens of thousands of tables are available without a master schema or explicit knowledge of how their contents relate to one another.
