SchemaLogix: Advancing Interoperability with Machine Learning in Schema Matching

Abstract

Schema matching, a fundamental process in data integration, traditionally employs pairwise comparisons to discern semantic correspondences among elements in disparate schemas. However, recent developments underscore the necessity of concurrent matching of interconnected schemas, termed schema alignment, to reconcile heterogeneous elements. This paper presents SchemaLogix, an innovative machine learningbased approach for schema matching. SchemaLogix addresses challenges such as data scarcity and domain-specific constraints through an inventive bootstrapping method, autonomously generating extensive datasets. Furthermore, SchemaLogix capitalizes on inherent alignment context constraints to optimize learning and improve precision across varied schema structures. Additionally, SchemaLogix incorporates user contributions to validate chosen correspondences, refining outputs based on valuable feedback. Empirical evaluations establish SchemaLogix's superiority over traditional methods, achieving an exceptional maximum S1 score of 0.90. These results offer practical insights for real-world applications, substantially advancing data integration and interoperability endeavors.