Website Fingerprinting Attack on Anonymity Networks Based on Profile Hidden Markov Model

Abstract

Website fingerprinting attacks can reveal the receiver in anonymous networks and cause a potential threat to users' privacy. Previous studies focus more on identifying individual webpages. They also neglect the hyperlink transition information, because it induces extra “noise” to classify the original webpage. However, it is a common scenario that the users surf a website by clicking hyperlinks on the webpage. In this paper, we propose a website modeling method based on profile hidden Markov model (PHMM) which is widely used in bioinformatics for DNA sequencing analysis. Our technique explicitly accounts for possible hyperlink transitions made by users when fingerprinting a target website, and therefore can work in a more realistic environment than existing methods. Using SSH and Shadowsocks, we collect various data sets and conduct extensive evaluations. We also show that our approach could work both in webpage and website identification in a closed world setting. The experimental results demonstrate that our website fingerprinting is more accurate and robust than existing methods.