An Ontology-based Mapping Repository for Dynamic and Customized Data Integration

There exist an increasing number of (semi-)structured information sources over the Web. Data integration techniques are often applied to increase the access efficiency of heterogeneous information sources. Data integration systems are built by combining a set of data sources for meeting specific user requirements. Due to various user needs, a growing trend of increasing variety of data integration system is noticeable even for the same application. In the academic community, much effort has been devoted to building a single data integration system. However, the existing approaches are impractical when the scale of data integration systems becomes large. This article proposes a key component of a total solution for holistically constructing a large number of customized data integration systems with the assistance of a considerable number of cooperative humans (a.k.a, a community). This component is an ontology-based mapping repository, called M-Ontology. Since each data integration system requires mappings for resolving semantic interoperability among information sources, MOntology aims at the efficient storage, management and discovery of mappings. M-Ontology is shared by all the data integration systems in the same application domain for mapping sharing and reuse. The internals of the mapping model are semantics-based for easy understanding and management by (even non-technical) community members. Following this model, a “humanfriendly” mapping insertion algorithm is proposed for construction of a well-organized mapping ontology through incremental insertion of domain-specific mapping instances. Three types of human intervention strategies (i.e., validation, avoidance and prevention) are designed for improving the performance of construction algorithms. In addition, this article presents a mapping discovery algorithm to find the many-to-many complex mappings by leveraging the mapping ontology. The discovered mappings are recycled to further enrich the mapping ontology using the insertion algorithm. Finally, we conduct two sets of experiments on real-world data integration over the Web, and the results show that the approach is feasible and effective.