Decentralised collaborative applications address privacy, availability and security issues related to centralised collaborative platforms. Such applications are based on a peer-to-peer communication paradigm according to which all users are directly connected to one another. Collaborations tend to widen and spread beyond the borders of organisations. Under these circumstances, it is necessary to guarantee to users the control over their data, while keeping collaboration available. To that end, the social network that has built between collaborators may be used as topology. Lack of information on this trusted network leads us to develop an approach to study its morphological properties.
In this thesis, we develop and implement an approach to study the social structure of interactions in the context of inter-organisational collaborations. We propose a stochastic approach based on \aclp{ERGM} and spatial models. We define a formalism that highlights the structure of interactions and integrates the organisational dimension. We propose to use a Bayesian inference method, ABC Shadow, to overcome the issues related to the parameters estimation. This approach is applied to a real case study: the collaborations initiated by researchers in a laboratory. In particular, it highlights the low tendency for a researcher to create collaborative links with other laboratories. We show that this approach can be applied to other kinds of social interactions, such as interactions between pupils of a primary school. Finally, we present a parallelisation strategy of the Gibbs sampler aimed at processing larger graphs in a reasonable time.
