Evolution dynamics of attribute-rich social networks

Abstract

We then explore the interdependence of opinion change and social group membership in human social networks. Previous work has shown persons are more likely to join groups when there is similarity in personal attributes and that people adjust their opinions to conform to the average of the groups they join.

We study the evolution dynamics in attribute-rich social networks. We start with demonstrating how node attributes can be used to predict the formation and dissolution of new links in these networks. We introduce a method for using node preferences for different attributes to predict link formation and dissolution. We then rank these attribute according to their importance for making predictions. We find that, in the university based social network that we study, personal preferences, in particular for political views and preferences for common activities help predict link formation and dissolution. We then we look at how link prediction can be used to identify changes in network stability. We demonstrate applications to collaboration networks. We also demonstrate how to identify large scale changes in link formation patterns using link prediction.

We then examine the dynamics of coevolution of three-layer node-aligned network: the first is defined by nominations based on perceived prominence collected from repeated surveys of students during their first four semesters of college; the second is a behavioral layer representing actual interactions between individuals based on records of their mobile calls and text messages; while the third is a behavioral layer representing potential face-to-face interactions implied by Bluetooth collocations. We investigate how these layers co-evolve: we study whether the formation or dissolution of a link in one of the layers precedes or succeeds formation or dissolution of the corresponding link in another layer (temporal dependencies), we explore the causes of observed temporal dependencies between the three layers and measure the predictive capacity of such dependencies; finally we observe the progress of nominations and the stages that lead to them.

We further explore the evolution of face-to-face interacting groups in social networks. Previous work has shown that selectivity based on opinions and values of attributes is an important tie-formation mechanism in human social networks. Less well-known is how selectivity influences the formation and composition of whole groups in which interactions extend beyond the dyads. To address this question, we use data from the NetSense study consisting of a multi-layer (nomination, communication, co-location) network of university students. We examine how group formation differs from tie-formation in terms of the role of selectivity based on opinions and attributes. In addition, we show how levels of such selectivity varies between groups formed to meet different needs.

Here we build on this work by looking at group joining and leaving and opinion change as a set of mutually reinforcing behaviors. We propose a benefit-driven mathematical model and a related machine learning based approach designed to predict the changes that people will make. We model benefit derived from a group membership as a combination of the amount of interactions the person has with the group and the level of opinion agreement with members. We use data from the NetSense study including records of evolving social groups and regularly collected opinions among university students. Our model is able to predict observed opinion changes in the data with relatively high accuracy. We also explore how having an opinion similar to the majority of members of the group affects the overall benefit a person's derives from social interaction thus allowing us to determine when they are likely to make a behavioral change.