Online Social Networks and Applications: a Measurement Perspective
Ben Y. Zhao
UC Santa Barbara
SWS Colloquium
Ben Zhao is a faculty member at the Computer Science department, U.C. Santa Barbara. Before UCSB, he completed his M.S.
and Ph.D. degrees in Computer Science at U.C. Berkeley, and his B.S. from Yale University. His research interests
include networking, security and privacy and distributed systems.
He is a recipient of the National Science Foundation's CAREER award, MIT Tech Review's TR-35 Award (Young Innovators
Under 35), and is one of ComputerWorld's Top 40 Technology Innovators.
With more than half a billion users worldwide, online social networks such as Facebook are popular platforms for
interaction, communication and collaboration between friends. Researchers have recently proposed an emerging class of
Internet applications that integrate relationships from social networks to improve security and performance. But can
these applications be effective in real life? And if so, how can we predict their effectiveness when they are deployed
on real social networks?
In this talk, we will describe recent research that tries to answer these questions using measurement-based studies of
online social networks and applications. Using measurements of a socially-enhanced web auction site, we show how
social networks can actually reduce fraud in online transactions. We then discuss the evaluation of social network
applications, and argue that existing methods using social graphs can produce to misleading results. We use results
from a large-scale study of the Facebook network to show that social graphs are insufficient models of user activity,
and propose the use of "interaction graphs" as a more accurate model. We construct interaction graphs from our Facebook
datasets, and use both types of graphs to validate two well-known social-based applications (Reliable Email and
SybilGuard). Our results reveal new insights into both systems and confirm our hypothesis that choosing the right graph
model significantly impacts predictions of application performance.