Identity Theft Detection in Mobile Social Networks Using Behavioral Semantics
Cheng Wang, Bo Yang, Jing Luo · 2017
User behavioral analysis is expected to be a key technique for identity theft detection in the Internet, especially in mobile social networks (MSNs). While traditional methods prefer to use explicit behaviors, a series of behaviors implicit in user's texts can probably provide much more accurate identity. And these implicit behaviors can be digged from texts by LDA. Besides the latent feature in texts, a behavior also include other features (e.g., spatial and temporal features). A joint feature including these features can be a better evidence for identity theft detection. In this paper, we use a probabilistic generative model to detect identity theft in MSNs. We are going to conduct experiments on two real-life datasets: Foursquare and Yelp. A early experiment shows that semantic features achieve better performance than spatial features and we are conducting our main experiment to see a better performance with joint behavioral feature.