Modeling anomalous attention over an online social network through read/post analytics

Abstract

Online social platforms revolutionarize the way in which people communicate, shattering physical boundaries and bringing people together in the virtual environment. While users are able to access information and share knowledge with unprecedented ease and openness, danger also lurks in the dark. Social networks have the potential to draw unwanted and anomalous attention to their users. Through online social networks, the daily routines of an individual may be under constant surveillance of others. Such risks are closely associated with information leakage, and have posed serious privacy and safety concerns. This paper investigates such risks, which are typically captured by excessive, unprecedented and persistent gathering of personal information through the cyberspace. We focus on ways to mitigate such risks through formalizing the concepts of anomalous attention. This is a challenging question, as such behaviors are usually victim-defined and often occurs without visible trace. Viewing a network as interconnected nodes who exchange information through posting and reading messages, we provide an abstract model of attention, and quantify the level of attention a user pays towards another. Analyzing the sequence of attention between pairs of users in the network allow one to capture anomalous activities.