The recent growth of the Linked Open Data (LOD) cloud allowed for a fast and heavy publishing of a huge quantity of linked and open datasets, notably in the shape of RDF graphs. Integrating datasets from the LOD cloud must enable the unified querying of different data sources, and also enrich the database of an organization, create augmented content for applications, or even provide a standardized formatted way for publishing proprietary data on the World Wide Web. However, the publication of structured datasets without any kind of control can lead to the leakage of sensitive information and this is the major factor hindering the sharing of data for numerous institutions, since they lack guarantees about their own information and regarding possible data cross-matching by an external attacker as well, even if datasets were "anonymized" locally beforehand (this type of attack is possible via entity resolution, an automated process enabling the matching of two identifiers to access the whole set of information regarding both targets). The goal of this Ph.D. is therefore to explore and design formal and algorithmical solutions to these problems, to define anonymization and entity resolution methods and check the compatibility of such rules with privacy nad utility policies, formulated beforehand by dataset providers.