Analysis of social networks constituted by criminal organizations can capitalize on knowledge about their structures to facilitate the detection of their key actors. However, investigative acts establish links of different natures (e.g., geolocation, ownership) between different types of entities (e.g., people, places, vehicles). Thus arises the challenge of extracting the social network from the multivariate investigation graph by considering all available information. It is then be possible to perform structural analyses based on centrality measures, supporting the identification of key actors in the network. This paper proposes a method for extracting a social network from such a multivariate graph, the variables attached to the nodes and edges of the multivariate graph being taken into account to quantify the likelihood of the links induced in the social network.