A framework for schema matcher composition

Enterprise schemas tend to be different, which is the key issue when the seamless communication between systems is of utmost importance. One solution could be the development of standards which then could be enforced, however, vendors seem to be reluctant to comply with them and communication between existing and legacy systems still remains unsolved. Other solution could be schema matching, which resolves the matter on data level and the process do not require vendors to adhere to any kind of predefined schemas. The task is very complex on the other hand, even for human evaluators. Some of the solutions aired so far are fairly promising, however, their accuracy varies. Our goal was to find means by which the results could be enhanced. We have been focusing on the development of solutions which do not change the concept of the algorithms, but fine-tune them so that they achieve higher accuracy. Our experiments showed that the results of the matchers may vary on a large scale depending on the actual parameter settings. It has also turned out that the parameters should set for each scenario individually, as the best results are warranted only this way. In this article, we present a general approach for optimally dissembling existing solutions, and combining some of the resulting components in a way that the new matcher supersedes the donor ones. The composition and the optimal parameter setting combined provide a framework, which is capable of an enhanced performance. Improved accuracy lessens the need for the follow-up human supervision. Key-Words: schema matching, optimization, algorithm analysis, performance improvement, framework definition