The design and implementation of FlexiMatch a learning , flexible & extendible framework for matching schemas

Automating matching schemas has been under investigation in many areas for already some decades, but matching schemas is still often done manually by domain experts. Due to the rapidly increasing number of heterogeneous and distributed data sources in enterprises and on the web, the manual matching approach is more and more a limitation and the need for automating the schema matching process is increasingly important. At the start of this report we investigate the different approaches in the schema matching world. From evaluations of implementations based on these approaches, schema matching 'trends' and problems are discerned, which are taken as the starting point of this thesis. This report describes the schema matching framework FlexiMatch, which supports the multi-strategy approach, with each strategy represented as a Validator. Key characteristics of FlexiMatch are that: • FlexiMatch and its Validator-components can learn from previous mappings. • Validators can easily be added to, or selected from the Validator repository, in order to boost future matching performance or to adapt the system to the match task at hand. The main perspective of FlexiMatch, is that the elements of schemas (relational column and table elements, or XML elements and attributes) from a certain domain share domain concepts. FlexiMatch learns these concepts from previous mappings. Schema elements belonging to a certain domain concept can have various representations. Within FlexiMatch concepts are therefore represented by interconnected subconcepts. Such an interconnected subconcept group is derived from the different schema elements representations of a certain domain concept which FlexiMatch encountered during previous schema mappings. These subconcepts and their interrelations are used as an intermediate schema to derive matches between input schema elements. Schema elements are therefore first matched with subconcepts. Schema elements that are matched with similar, interrelated subconcepts are then combined with each other. The main goal of the FlexiMatch system for this thesis is to make a learning framework in which the gained knowledge could be used in future match tasks. Although there are many aspects of FlexiMatch that could be improved or implement to enhance the performance, the evaluation of FlexiMatch shows that the main goal is achieved: subconcepts and subconcept relations are learned from previous mappings, and they help in finding new ones in later match tasks.