Publications (1)0 Total impact
ABSTRACT: We take schema mapping to be the problem of finding an appropriate semantic relationship to load data from a source to a target database, given their schemas. This rela-tionship is expressed in terms of declarative logical expres-sions. The problem is inherently difficult to automate and previous solutions have proposed algorithms which take as input simple element correspondences between schemas in addition to local schema constraints such as keys and ref-erential integrity. In this paper, we investigate the use of a richer source of information about schemas, namely the presumed presence of semantics for each table expressed in terms of a conceptual model (CM) associated with it. Our approach first compiles each CM into a graph and repre-sents each table's semantics as a subtree in it. Second, we develop algorithms for discovering subgraphs that are plausible connections between those concepts/nodes in the CM graph that have attributes participating in element cor-respondences. A conceptual mapping candidate is then a pair of source and target subgraphs which are semantically similar. At the end, these must be converted to database mappings. We offer experimental results demonstrating that, for test datasets drawn from a number of domains, the "semantic" approach outperforms the chase technique which only uses referential integrity constraints, in terms of recall and especially selectivity.