Loading…
This event has ended. View the official site or create your own event → Check it out
This event has ended. Create your own
View analytic
Wednesday, July 13 • 15:31 - 17:00
Modeling misclassifications in multilayer networks

Sign up or log in to save this to your schedule and see who's attending!

Location: PSH (Professor Stuart Hall Building) - LG01, 
Goldsmiths, University of London, Building 2
Campus Map 

Contributor: Devin Gaffney, Northeastern, United States

Background: 

Haythornthwaite and Wellman’s seminal work establishing the importance of multiplexity in social contacts across various communication media established from an early point that accounting for the various networks individuals interact upon is of primary importance. This work addresses a methodological concern with studies on multiplexity - specifically, this work imagines trials of synthetic multilayer networks where ties made across layers are potentially incorrect in several ways. The work then examines the effect of these incorrect ties in terms of how analyzing the diffusion of a rumor may differ from cases where all ties are correctly assigned.


Objective: 

This paper aims to contribute to methodological practices around measuring multilayer networks in emergent situations, such as a breaking news event of an online activist campaign. The issue at hand is largely concerned with failures to correctly identify cross ties between network layers, and how various failures result in different outcomes than an identical counter-factual case where those failures are not present.


Methods: 

The work employs a network modeling approach combined with a rumor diffusion model. The paper establishes several parameters of interest that approximate different types of failures in generating ties between two networks, and then modulates those parameters randomly over many stochastic realizations of the model, while always having two control cases with the same parameters for each realization to allow for a comparison between what is different in a case where failures occur. From this, a comparison one zero failure (where ties across networks are perfectly set) case against the other zero failure case and one zero failure case and the failure case (where ties across networks are imperfectly set) allows for a close examination of the impact of these failures on being able to correctly measure outcomes from the network.


Results: 

The results indicate that only certain types of errors actually damage downstream analysis of emergent events in multilayer networks. Specifically, as long as the approximate number of cross ties is close to the correct amount, even if those ties are misclassified, the results will mostly allow for a close analysis that is correct in its findings. If, however, many ties fail in the sense that those links between the networks are never drawn, the results will deviate from the control case.


Future Work: 

Future work will consider further complications arising in multilayer network situations, such as differing parameters for rumor spreading (i.e. a condition where one network spreads the rumor differently than the other) while also removing the complexity of the current work for parameters that seem to have little to no effect on differing outcomes. 

Wednesday July 13, 2016 15:31 - 17:00
PSH (Professor Stuart Hall Building) - LG01 Goldsmiths University, Building 2

Attendees (8)