Smart Matching – An Approach for the Automatic Generation of Executable Schema Mappings

This work has been finished in September 2008.

Information integration has a long history in computer science [1]. It has started with theintegration of database schemas in the early eighties. With the rise of the Semantic Web andthe emerging abundance of ontologies, the need for an automatic information integration increased further.

Information integration in general and automatic information integration in particular is a huge and challenging research area. One of the main problems is handling semantic heterogeneity and schema heterogeneity. Manually finding the semantically overlapping parts of schemas is a tedious problem. Furthermore, writing integration code is a labor intensive, error-prone, and cumbersome task. A lot of approaches have already been developed to automate this work. Nevertheless, not all integration problems have been solved so far.

Matching tools are used to automatically find similarities between schemas. The results of these tools are simple correspondences. Based on these correspondences, one is able to write integration code. However, the simple correspondences are just suggestions and must be verified manually. Hence, the completeness and correctness of the resulting correspondences may not be assured. Furthermore, it is not possible to automatically derive transformation code for all found simple correspondences.

In order to write transformation code, different kinds of transformation languages have been developed. The produced code is too customized for a specific type of schema to be easily reused for other integration problems. Hence, to the best of our knowledge, there exists no transformation language to develop reusable transformation patterns for different kinds of heterogeneity problems.

This thesis addresses the heterogeneity problems, as well as the lack of reusable transformation code, and the need for establishing correct and complete correspondences between schemas. The first two problems are tackled by developing an executable declarative mapping language, which is able to cope with the core of schema heterogeneity problems. In contrast to simple correspondences, this mapping language is able to express more constraints. Based on these more expressive mappings, the execution code is automatically derived. The third problem is tackled by a self-tuning, iterative matching approach. This approach is based on the developed mapping language. Mapping strategies are responsible for the application of mapping operators. Based on the executable mapping suggestion, completeness and correctness are achieved for a provided set of instance models by a test-driven approach. These instance models are used to evaluate the produced mapping model. The prototype of this self-tuning approach is called SmartMatcher.

[1] Laura Haas. Beauty and the Beast: The Theory and Practice of Information Integration. In 11th International Conference on Database Theory, Springer LNCS 4353, 2007, pp. 28-43.

Download the paper