Estimação em modelos de redes de afinidade

Abstract

The analysis of social networks aims to detect and measure relationships between elements in the most diverse sectors of society. Therefore, numerous methods are used for this type of analysis, among which we highlight the affinity networks model. This model of random graphs considers that the probability of connection is a function of characteristics of features and this allows a better approximation of the model to real networks. The present work proposes to produce a methodology to perform the estimation of the parameters that compose the model of affinity networks. In case there is only one community, we use the likelihood function directly to nd the estimators, in case there is more than one community, we build an Expectation Maximization algorithm, as we have a latent model. We ensure that the method works well through simulations. We propose a method to determine the appropriate number of communities in the affinity network and allocate individuals to them. Graphs were made to represent affinity networks using two different methods to measure affinity between individuals in the network. Finally, we applied the methodology presented in a real situation, in a data collection in order to describe how the people interviewed had their lives impacted by the Covid-19 pandemic.