Utility-based model for characterizing the evolution of social networks

Characterizing the evolution of social networks is challenging, because social networks tend to be large. In addition, their evolution entails latent variables, such as unobserved meeting states. The objective of this paper was to develop a utility-based model with preference parameters for clarifying the reasons for the behavior underlying network evolution. The expectation-maximization (EM) algorithm combined with snowball sampling (SS) was proposed for parameter estimation. EM is suitable to cope with the data set with latent variables or missing data and SS would be helpful in decreasing the network size without loss of critical information. Simulation studies were further performed to assess the effects of network size, seed set size, meeting probabilities, sampling waves, and sampling methods on the summary statistics of estimates and to highlight the importance of the unobserved variables. Finally, an application was developed on the Facebook platform to demonstrate the model. In all, the proposed model is expected to be a novel tool for exploring and understanding the evolution of social networks.