PRISMA: A Privacy-Preserving Schema Matcher using Functional Dependencies

Schema matching aims to find semantic correspondences between the columns of two schemas. Due to its high relevance in the field of data integration, it has been extensively studied in the literature. However, most matching approaches assume similarity of column names or instance data of the schemas to be matched and struggle when these are encoded differently, e.g., if they have been encrypted due to privacy requirements. We present Prisma, a novel encoding-independent schema matcher that utilizes functional dependencies to construct graph embeddings that exploit the encoding-independent structure of the schemas to be compared. We compare Prisma against multiple baseline matchers as well as state-of-the-art competitors. The experiments demonstrate that Prisma outperforms these approaches on databases that have large differences in their encodings, especially if these databases consist of multiple tables.