Pay-as-you-go ranking of schema mappings using query logs

Abstract

Data integration systems typically make use of mappings to capture the relationships between the data resources to be integrated and the integrated representations presented to users. Manual development and maintenance of such mappings is time consuming and thus costly. Pay-as-you-go approaches to data integration support automatic construction of initial mappings, which are generally of rather poor quality, for refinement in the light of user feedback. However, automatic approaches that produce these mappings typically lead to the generation of multiple, overlapping candidate mappings. To present the most relevant set of results to user queries, the mappings have to be ranked. We proposed a ranking technique that uses information from query logs to discriminate among candidate mappings. The technique is evaluated in terms of how quickly stable rankings can be produced, and to investigate how the rankings track query patterns that are skewed towards specific sources. © 2012 Springer-Verlag.