Morpho-statistical description of networks through graph modelling and Bayesian inference

Quentin Laporte-Chabasse (1) , Marianne Clausel (2) , Radu Stoica (2) , François Charoy (1) , Gérald Oster (1)

1 COAST - Web Scale Trustworthy Collaborative Service Systems
2 IECL - Institut Élie Cartan de Lorraine

Collaborative graphs are relevant sources of information to understand behavioural tendencies of groups of individuals. Exponential Random Graph Models (ERGMs) are commonly used to analyze such social processes including dependencies between members of the group. Our approach considers a modified version of ERGMs, modeling the problem as an edge labelling one. The main difficulty is inference since the normalising constant involved in classical Markov Chain Monte Carlo (MCMC) approaches is not available in an analytic closed form. The main contribution is to use the recent ABC Shadow algorithm. This algorithm is built to sample from posterior distributions while avoiding the previously mentioned drawback. The proposed method is illustrated on real data sets provided by the Hal platform and provides new insights on self-organised collaborations among researchers.

Domaines: Réseaux sociaux et d'information [cs.SI]; Statistiques [math.ST]

Quentin Laporte-Chabasse, Marianne Clausel, Radu Stoica, François Charoy, Gérald Oster. Morpho-statistical description of networks through graph modelling and Bayesian inference. 2019. ⟨hal-02421787v1⟩