Résumé
This paper deals with redundant data, previously identified by a data linkage step. The question considered is: how to propose a unique representation by merging the identified "duplicates"? More specifically, how to decide, for each data property, which value will be chosen among those describing the "duplicates"? What method should be adopted in order to be able to trace and explain the result to the user in an understandable form? The proposed approach relies both on multi-criteria decision and argumentation, combined with the computation of a quality score.