Ruffled feathers: trait inference beyond homophily

Co-Sponsored by the Computational Social Science (CSS) Initiative through the Center for the Study of Complex Systems
Abstract:
The observation that individuals tend to be friends with people who are similar to themselves, commonly known as homophily, is a prominent and well-studied feature of social networks. In the presence of homophily, attributes of one's friends are predictive of one's own attributes, which is classically exploited by methods aimed at predicting missing attributes. While weak homophily might suggest that attribute prediction is difficult, in this work we find a common pattern of extreme preferences for attributes that introduce friend-of-friend or two-hop correlations, where individuals are similar to their friends-of-friends without necessarily being similar to their friends. We call this property monophily for ``love of one.'' We contribute a joint characterization of the statistical structure of homophily and monophily in terms of preference bias and preference variance and demonstrate how homophily-based prediction methods based on friends, ``the company you keep,'' are fundamentally different from monophily-based methods based on friends-of-friends, ``the company you're kept in.'' As part of this work we contribute a new stylized model of social network structure that we call the overdispersed stochastic block model, where the intensity of homophily and monophily can be independently modified to simulate networks with realistic preference bias and variance. To illustrate the differences between homophily and monophily-based prediction, we place particular focus on gender prediction on social networks, where homophily can be weak or nonexistent in practice. Yet we also demonstrate that in settings where homophily is present, such as the link structure of political blogs, monophily is also present and can power methods that outperform homophily-only methods for prediction. These findings offer an alternative perspective on attribute prediction in general and gender in particular, complicating the already difficult task of protecting attribute privacy on social networks.

Bio:
Johan Ugander is an Assistant Professor at Stanford University in the Department of Management Science & Engineering, within the School of Engineering. His research develops algorithmic and statistical frameworks for analyzing social networks, social systems, and other large-scale social data. Prior to joining the Stanford faculty he was a post-doctoral researcher at Microsoft Research Redmond 2014-2015 and held an affiliation with the Facebook Data Science team 2010-2014. He obtained his Ph.D. in Applied Mathematics from Cornell University in 2014. His awards include a Best Paper Award (2012 ACM WebSci), a Best Student Paper Award (2013 ACM WSDM), and the 2016 Eugene L. Grant Undergraduate Teaching Award from the Department of Management Science & Engineering.