Predictive Analysis on Cyber-predators Using Various Social Network Structures

Abstract

In this paper, we use the information of social network structures to tackle cyber-predator detection in a social networking service, and compare and analysis the explanatory power of these structures. We first create networks from various perspectives such as footprints and reactions to posts as well as conversations. By applying Large-scale Information Network Embedding (LINE) to these networks, latent representations based on each network structure are extracted. Using these latent representations as input features, we develop classification models for predicting cyber-predators. As a result of computational experiments, we confirmed that many social network structures are effective for detecting cyber-predators. In addition, we got some interesting findings, such as "the tendency of cyber-predator appears most strongly in the profile browsing history". The findings obtained in this paper are used to suppress minors' cybercrime damage.