Repositori URV

Identificador: imarina:9298244

Handle: https://hdl.handle.net/20.500.11797/imarina9298244

Autors:
Martinez, SergioSanchez, DavidValls, Aida

Resum:
Record linkage methods evaluate the disclosure risk of revealing confidential information in anonymized datasets that are publicly distributed. Concretely, they measure the capacity of an intruder to link records in the original dataset with those in the masked one. In the past, masking and record linkage methods have been developed focused on numerical or ordinal data. Recently, motivated by the proliferation of textual information, some authors have proposed masking methods to anonymize textual data. Textual attributes should be interpreted according to their semantics, which makes them more difficult to manage and compare than numerical data. In this paper, we propose a new record linkage method specially tailored to accurately evaluate their disclosure risk. Our method, named Semantic Record Linkage, relies on the theory of semantic similarity and uses widely available ontologies to interpret the semantics of data and propose coherent record linkages. Test performed over a real dataset shows that a semantic record linkage method evaluates better the disclosure risk when compared with a non-semantic approach.