Inteplato: Generating Mappings of Heterogeneous Relational Schemas Using Unsupervised Learning
Leonard Traeger, Andreas Behrend, George Karabatis · 2022
The growth of unstructured, unclean, and incomplete data poses quite a problem on attempts to integrate information from various disparate repositories. This paper describes Inteplato, a scalable web-based framework for domain-independent schema mappings between numerous heterogeneous databases. Starting with schema XML exports, the local schemas are visualized in a knowledge graph. A novel algorithm generates clusters based on similarity of local concepts by utilizing fuzzy string, synonym intersection, data type, and constraint similarities. Using these clusters, another algorithm generates mappings between global and local level concepts. The autonomy of each participating database system is maintained, and the mappings enable code automation for global-to-local query propagation. Accuracy was used as a metric to evaluate the generated mappings between local and global concepts.