Selection and Configuration of Schema Matchers

Schema matchers use a wide variety of information and methods to identify correspondences in two given schemas. A simple string comparison between the labels of schema elements is sometimes sufficient to achieve a good result. Often, additional information has to be taken into account, e.g. the schema structure, background knowledge such as thesauri, ontologies or existing mappings. Even if just a simple string matching technique is used, the string matcher has still to be configured with the right parameters (an important aspect is, for example, the tokenization of labels).

Therefore, the selection and configuration of schema matchers is a challenging problem. People experienced in schema matching and having knowledge about the schemas to be matched still have to invest some time in order to select the right methods and parameters for a given matching problem.

The goal of this thesis is to develop mechanisms to automatically select and configure matchers for a matching problem. These mechanisms should be based on the characteristics of the input schemas, such as labels, structure, and availability of (domain-specific) background knowledge. These characteristics should be represented in a feature vector, and a configuration which is known to work well for problems with similar feature vectors is chosen.

The mechanisms to select and configure schema matchers automatically should be implemented in the matching framework of our generic model management system GeRoMeSuite. GeRoMeSuite offers already various matchers, which can be combined in a very flexible and extensible matching framework. The evaluation should use data sets from existing matching tasks, such as OAEI (Ontology Alignment Evaluation Initiative) or other schema matching literature.