I Know What You Saw Last Minute - Encrypted HTTP Adaptive Video Streaming Title Classification

摘要: Desktops and laptops can be maliciously exploited to violate privacy. There are two main types of attack scenarios: active passive. In this paper, we consider the passive scenario where adversary does not interact actively with device, but he is able eavesdrop on network traffic device from side. Most Internet encrypted thus attacks challenging. Previous research has shown that information extracted multimedia streams. This includes video title classification non HTTP adaptive streams (non-HAS). paper presents an algorithm for streaming classification. We show external attacker identify (HAS) sites such as YouTube. To best our knowledge, first work shows this. provide a large data set 10000 YouTube 100 popular titles (each downloaded times) examples task. The dataset was collected under real-world conditions. present several machine algorithms task run through experiments, which accuracy more than 95%. also classify in training unknown some eliminate false prediction instead report unknown. Finally, evaluate robustness delays packet losses at test time solution uses SVM most robust against these changes given enough data. crawler future research.