Schema Meta - Matching Extended

Schema matching, the process of matching between concepts describing the meaning of data in heterogeneous, distributed data sources (e.g. database schemata, XML DTDs, HTML form tags, etc.) is one of the basic operations required by the process of data integration. Recently, several algorithms for automatic schema matching have been proposed and evaluated in the database community. While in many domains these tools succeed in finding the right matching, empirical analysis shows that there is not (and probably will never be) any single algorithm that is guaranteed to succeed in all possible domains and applications. To overcome this problem, several tools are being developed that combine the principles by which different algorithms judge the similarity between concepts. In parallel, Anaby-Tavor et al [1] takes another approach, by which not one, but K best-ranked mappings are generated, then examined iteratively until a good mapping is found. In this paper we introduce a novel framework for schema matching which we call schema meta-matching. This approach extends the idea of working with top-K mappings, applying it to an arbitrary ensemble of algorithms for schema matching. Informally, schema metamatching is the problem of computing a “consensus” ranking of alternative mappings between two sets of concepts, given the “individual” graded rankings provided by several algorithms for schema matching. We begin with an overall look at schema matching, concluding with a discussion of how the field is likely to develop in the near future. We formalize the problem of schema meta-matching and introduce several algorithmic solutions for this problem, including one that adapts standard techniques for general quantitative rank aggregation, and others employing novel techniques specific to the problem of schema matching. We provide a formal analysis of the applicability and relative performance of each competing algorithm. Finally, we show how combining these approaches results in the most successful matchings.