The emergence of big data, as well as open data, has led the scientific community to question data collection in different ways: construction, acquisition and ownership. Traditionally, social sciences have been concerned by the opportunities given by data collection or access to existing datasets. The recent changes related to automatic data collection and sharing makes attitudes and expectations of researchers evolve. Social sciences are thus expected to change their current approaches and methodologies regarding data. Geography is a discipline largely concerned with data collection and analysis and is then at the core of this dramatic change. The wide and fast spread of numerous geolocation tools (based on GPS, GSM, WiFi or IP address) leads to the surge of georeferenced data (Audard, Carpentier, Oliveau, 2014). If an enthusiastic posture currently dominate the appreciation of open/big data in terms of scientific perspectives (Marx, 2013), some cautions emerges regarding their use. These data are generally not built for the purpose of social sciences (Pumain, 2014) and are thus to be used with care for scientific work (Terrier, 2011). Beyond those general considerations, one major fact related to their use in geography is the question of the spatial, social, demographic and economic representativeness of such « second hand » datasets built for their own purpose and objectives. In the case of big data, the illusion of completeness tends to obscure the question of representativeness. The huge number of observations does not mean that there is no selection bias; even if a big sample size produce good confidence intervals. That's the main question of this paper: how to reshape the socio-spatial representativeness from data that are not built with regard to the question of spatial representativeness?