Utilizing Survival Analysis for Modeling Child Hazards of Social Networking

Social networks induce several hazards to children, which are correlated with the amount of time that children are exposed to those networks. To this end, this work investigates the relation of the aforementioned hazards with the exposure time. To address this issue, we adopt techniques used in survival analysis. These techniques involve the estimation of certain functions which reflect the relation of a disastrous event with time. In particular, we derive the distribution of the rate at which suspicious activities towards children occur in social networks. This is accomplished through experiments on data sets extracted from Facebook. The results show that the incoming hazards for minor female profiles follow the Logistic distribution, while the corresponding hazards for minor male profiles follow the Normal distribution. This knowledge is then utilized for developing an effective system for automated grooming recognition, by optimizing the detection threshold as a function of time. Thus, the threshold sensitivity can be appropriately adjusted such that lower frequencies of occurrence lead to lower threshold sensitivities, and higher frequencies of occurrence lead to higher threshold sensitivities.