Predicting offline behaviors from online features

Abstract

Investigating online social behaviors may help us to better understand and predict offline high risk behaviors in gay communities. But how can offline behaviors be predicted from online social networks? This article selects data from 26 online social network groups from QQ (a Chinese based messaging software) administered by gay communities of "W" city of Hubei Province, China. Based on online data mining, social network analysis, and offline semi-structural interviews, we argue that the ego-centric dynamical network analysis---an approach which combines partial network dynamics, individual features, and structure position together---can be used to derive the probabilistic features for predicting offline high risk behaviors (HRB). An example of HRB is "one night stands" (gays for one night: 419) for gay homosexuals.