CoEGSS (Center of Excellence for Global System Science) [1] is a center of excellence of the European Union for enhancing the power of Global System Science simulations with the calculation power of High Performance Computing. Three pilots are studied: the diffusion of health habits, the attitude to green mobility and urbanization. In the case of social simulations, it is crucial to properly model the social ties among individuals. Data about the real population are aggregated, due to privacy constraints. Thus, in order to have a description per agent, a synthetic population is created to meet the aggregated constraints. Agents are described by a sequence of attributes of different type, as for instance “sex” (a categorical entry), “age” (an ordinal value) or “household location” (usually latitude and longitude). In our approach we infer social relations from the similarity among agents, by extending Lin similarity [2] to different type dependent attributes. The idea is to consider the information shared by each couple of agents. We will show results applied on the synthetic population of CoeGSS as well as on the (real) dataset of leather companies in Tuscany and on the Italian National dataset of running cars.