Hierarchical sequencing of online social graphs

TL;DR

This paper analyzes the topology of online social networks, especially MySpace dialogs, using simplicial complexes to identify community structures, emotion propagation layers, and a new measure of social capital called topological dimension.

Contribution

It introduces a novel topological analysis framework for online social graphs, including the concept of node's structure vector and topological dimension, to quantify social capital and emotional dynamics.

Findings

Network exhibits multilayered community structures.

Emotion spreading occurs in two distinct layers with different topologies.

Node's topological dimension correlates with social brokerage ability.

Abstract

In online communications, patterns of conduct of individual actors and use of emotions in the process can lead to a complex social graph exhibiting multilayered structure and mesoscopic communities. Using simplicial complexes representation of graphs, we investigate in-depth topology of online social network which is based on MySpace dialogs. The network exhibits original community structure. In addition, we simulate emotion spreading in this network that enables to identify two emotion-propagating layers. The analysis resulting in three structure vectors quantifies the graph's architecture at different topology levels. Notably, structures emerging through shared links, triangles and tetrahedral faces, frequently occur and range from tree-like to maximal 5-cliques and their respective complexes. On the other hand, the structures which spread only negative or only positive emotion messages appear to have much simpler topology consisting of links and triangles. Furthermore, we introduce the node's structure vector which represents the number of simplices at each topology level in which the node resides. The total number of such simplices determines what we define as the node's topological dimension. The presented results suggest that the node's topological dimension provides a suitable measure of the social capital which measures the agent's ability to act as a broker in compact communities, the so called Simmelian brokerage. We also generalize the results to a wider class of computer-generated networks. Investigating components of the node's vector over network layers reveals that same nodes develop different socio-emotional relations and that the influential nodes build social capital by combining their connections in different layers.