SmartMatcher -- How Examples and a Dedicated Mapping Language can Improve the Quality of Automatic Matching Approaches

Information integration has a long history in computer science. It has started with the integration of database schemas in the early eighties. With the rise of the semantic Web and the emerging abundance of ontologies, the need for an automated integration increased further. A lot of automated matching approaches and tools have been proposed so far. The typical output of such tools is a simple one-to-one alignment mostly based on schema information, e.g., similar names and structures of schema elements. However, these alignments cannot cope with schema heterogeneities, hence, these problems must be resolved manually. Furthermore, there is no automated evaluation of the quality of the alignments based on the instance level, because the matching approaches are not bound to a specific integration scenario, e.g., transformation or merge. In this work we propose the SmartMatching approach, which can be seen as an orthogonal extension to existing matching approaches for increasing the quality of the automatically produced alignments for the transformation scenario. This is achieved by using an executable mapping language for bridging schema heterogeneities and by using instance models to evaluate the quality of the alignments in an iterative and feedback-driven process inspired by machine learning approaches.