Measuring Influence in Social Networks Through Information Diffusion Modeling

• Activity Number: 568
• Type: Contributed
• Date/Time: Wednesday, August 6, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Donggeng Xia*+ and George Michailidis
• Companies: and University of Michigan
• Keywords: Cumulative Cox hazard rates ; user influence ; Rank prestige algorithm

Abstract

Data extracted from social network communication platforms, such as Twitter, records users' interactions over time. A key question is to determine who are the most influential members in such networks. A common influence measure discussed in the literature is based on a variant of the popular Page-Rank algorithm. In this talk, we propose a modification of such rank prestige algorithms by modeling the weights used to reflect users' activity, as opposed to users connectivity that is currently the case. The users activity is captured through interactions between users on the information they post, rebroadcast or comment on. These activities are modeled as multivariate interacting counting processes. We discuss how to estimate their parameters through maximum likelihood and establish their asymptotic properties. The proposed model is illustrated on simulated data as well as a data set extracted from discussions on Twitter.