|Authors||D. T. Schroeder, P. Lind, K. Pogorelov and J. Langguth|
|Title||A Framework for Interaction-based Propagation Analysis in Online Social Networks|
|Project(s)||UMOD: Understanding and Monitoring Digital Wildfires, Department of High Performance Computing|
|Publication Type||Proceedings, refereed|
|Year of Publication||2020|
|Conference Name||Complex Networks|
Online social networks create a digital footprint of human interaction naturally by the way they function. Thus, they allow a large-scale analysis of human behavior which was previously infeasible for social scientists. Consequently, social networks have been studied intensely in the last decade. The core of most social networks is the relationship between users which can be described as a graph. The graph can be either undirected, as is the case for the friendship relation of Facebook, or directed, which is the case of the follower relation on Twitter. The relationship is readily visible, e.g. on the user interface of the social networks themselves. However, these edges are unweighted expressions of interest and reflect how individuals have chosen to relate to each other rather than how they actually interact with each other. For studying information propagation, comparing interaction properties is crucial and, therefore, using models based on connections that reflect different dimensions and strengths of acquaintance seems appropriate. Thus, there is a need for obtaining weighted edges from the communication that occurs on the social network. In this paper, we present a novel method to calculate an acquaintance score between pairs of Twitter users and use the resulting networks to enable the analysis of interaction based information propagation. By understanding the frequency and velocity with which individuals share content as a measure of acquaintance, it becomes possible to predict, compare communication patterns, and detect unusual communication. In contrast to previous work which assigns edge weights based on tie strength, our score considers the response time as a crucial factor and, therefore, enables time-based spreading comparisons.