Triadic Closure and Its Influence in Social Networks
A Microscopic View of Network Structural Dynamics
by Hong Huang
Date of Examination:2016-09-01
Date of issue:2016-09-12
Advisor:Prof. Dr. Xiaoming Fu
Referee:Prof. Dr. Xiaoming Fu
Referee:Prof. Dr. Marcus Baum
Referee:Prof. Dr. Wolfgang Nejdl
Persistent Address:
http://dx.doi.org/10.53846/goediss-5854
Files in this item
Name:thesis.pdf
Size:1.68Mb
Format:PDF
Description:final thesis for Hong
The following license files are associated with this item:
Abstract
English
Social triad—a group of three people—is one of the simplest and most fundamental social groups, which serves as the basis of social network analysis. Triadic closure, a closing process of an open triad, is a useful principle and model to understand and predict network evolution and community growth, which has been widely used in web mining and solving social issues like political movements, professional organizations and religious denominations. Extensive network and social theories have been developed to understand the triadic structure, for example, triadic closure facilitates cooperative behavior and "friend of my friends are my friends". However, over the course of a triadic closure—the transition from open triads to closed triads are much less well understood. Furthermore, the interaction dynamics in networks, particularly in a triad is still unclear. In order to fill the gap in triadic closure studies, in this thesis, we trace the whole process during and after triadic closure. Starting from open triads, we study the problem of group formation in online social networks and try to understand how closed triads are formed from open triads in dynamic networks. Secondly, we focus on triadic closure’s influence on networks, especially its influence on tie strength dynamics of social relations. We investigate whether the new established third link will affect the tie strength dynamics of open triads after triadic closure. Employing a large microblogging network as the source in our study, we first focus on open triads closing process. By investigating the impact of different factors from three aspects: user demographics, network characteristics, and social perspectives, we find some interesting phenomena including: male, celebrity and gregarious users are more inclined to closing triads; structural hole spanners are eager to close open triads for more social resources, but they are also reluctant to have two disconnected friends to be linked together. Then, we examine triadic closure and its influence – tie strength dynamics of triads after closure, especially whether and how the formation of the third tie among three users in a triad affects the strengths of the existing two ties using two dynamic networks from Weibo and mobile communication. We find that the closure of 80% social triads weakens the strength of the first two ties. Surprisingly, we discover that although males are easier to get closed, the decrease in tie strength among three males is more sharp than that among females, and celebrities are more willing to form triadic closure. However, the tie strengths between celebrities are more likely to be weakened as the closure of a triad than those between ordinary people. We also demonstrate that while strong ties result in weakened relationships in open triads, they can promote the stronger ties in closed social triads. Further, we formalize a prediction problem to predict triadic closure. We propose a probabilistic graphical method to solve the triadic closure prediction problem by incorporating user demographics, network topology, and social information. With better instantiating attribute factors, we also extended our model with kernel density estimates. Unlike triadic closure prediction, the prediction for triadic tie strength dynamics is far more complicated when time dynamics is took into account. We further propose a dynamic probabilistic graphical to solve the problem of triadic tie strength dynamics prediction with the consideration of user demographics and temporal as well as structural correlations. Extensive experiments demonstrate that our proposed model offers a greater predictability for both prediction tasks. We demonstrate that our methodology offers a better-than-82% potential predictability for inferring the dynamics status of social triads in both networks, and the leveraging of the kernel density estimate together with structural correlations enables our models to outperform baselines by up to 30% in terms of F1-score. The triadic closure and its influence studied in this thesis will be a good guide to practical applications, like friend recommendation and new friend invitation for online microblogging services.
Keywords: Triadic Closure; Social Networks; Social Influence; Predictive Model
