Supervised learning for record linkage through weighted means and OWA operators

Record linkage is a technique used to link records from one database with records from another database, making reference to the same individuals. Although it is normally used in database integration, it is also frequently applied in the context of data privacy. Distance-based record linkage permits linking records by their closeness. In this paper we propose a supervised approach for linking records with numerical attributes. We provide two different approaches, one based on the weighted mean and another on the OWA operator. The parameterization in both cases is determined as an optimization problem. We evaluate our proposal and compare it with standard distance based record linkage, which does not rely on the parameterization of the distance functions. To that end we test the proposal in the context of data privacy by linking a data file with its corresponding protected version.