Statistical models of network structure are widely used in network science to reason about the properties of complex systems—where the nodes represent entities (e.g., users) and the links represent relationships (e.g., friendships). The goal is to develop generative network models (GNMs) that accurately capture the observed characteristics in real world networks to acquire a better understanding of the underlying properties of the system (e.g., social network). This is because a good descriptive model of social networks may not necessarily facilitate sampling. Along this area of research, it is also important to develop sampling methods for attributed networks (networks with vertex-attributes). However, this task remains a challenging problem because most current methods work with relatively simple GNMs.
In this talk, I will discuss the case of a number of recent GNMs that share a common form. This structure allows to increase both the variance of the models and the space of characteristics possible to be modeled. The structure also allows for a unified efficient sampling method. However, since in this structure the probability mass is allocated to certain regions of the network space it is hard to identify candidate networks to sample attributed networks. To solve this problem I will talk about a novel sampling method, CSAG, that generates attributed networks from GNMs equivalent to the new representation. CSAG constrains every step of the sampling process to bias the search to regions of the space with higher likelihood. The results show that CSAG jointly models the correlation and structure of the datasets better than the state of the art, and maintains the variability of the GNM while providing a reduction of 5 times or more in the error of the correlation.
Just before the start time, go to the event URL below and log on as "guest" with your name.
Event Link: https://csoi.adobeconnect.com/education