Generating SPARQL Executable Mappings to Integrate Ontologies

Abstract

Data translation is an integration task that aims at populat- ing a target model with data of a source model by means of mappings. Generating them automatically is appealing insofar it may reduce inte- gration costs. Matching techniques automatically generate uninterpreted mappings, a.k.a. correspondences, that must be interpreted to perform the data translation task. Other techniques automatically generate ex- ecutable mappings, which encode an interpretation of these correspon- dences in a given query language. Unfortunately, current techniques to automatically generate executable mappings are based on instance ex- amples of the target model, which usually contains no data, or based on nested relational models, which cannot be straightforwardly applied to semantic-web ontologies. In this paper, we present a technique to auto- matically generate SPARQL executable mappings between OWL ontolo- gies. The original contributions of our technique are as follows: 1) it is not based on instance examples but on restrictions and correspondences, 2) we have devised an algorithm to make restrictions and correspondences explicit over a number of language-independent executable mappings, and 3) we have devised an algorithm to transform language-independent into SPARQL executable mappings. Finally, we evaluate our technique over ten scenarios and check that the interpretation of correspondences that it assumes is coherent with the expected results.