Casamento de esquemas de banco de dados aplicando aprendizado ativo

Abstract

Given two database schemas within the same domain, the schema matching problem is the task of finding pairs of schema elements that have the same semantics for that domain. Usually, this task was performed manually by a specialist making it tedious and costly because the specialist should know the schemas and their domain. Currently this process is assisted by semi-automatic schema matching methods. Current, methods use some heuristics to generate matchings and many of them share a common modeling: they build a similarity matrix between the elements from functions called matchers and, based on the matrix values, decide according to a criterion which of the matchings are correct. This thesis presents an active-learning based method that uses the similarity matrix generated by the matchers, a machine learning algorithm and specialist interventions to generate matchings. The presented method di↵ers from others because it has no fixed heuristic and uses the specialist expertise only when necessary. In our experiments, we evaluate the proposed method against a baseline on two datasets: the first one was the same used by the baseline and the second containing schemas of a benchmark for schema integration. We show that baseline achieves good results on its original dataset, but its fixed strategy is not as e↵ective for other schemas. Moreover, the proposed method based on active learning is shown more consistent achieving, on average, F-measure value of 0.64.