Topo Noticias
QTDEI - CHARACTERIZATION AND DETECTION OF USER BEHAVIOR IN ONLINE AND MOBILE RANDOM VIDEO CHAT [2015-07-15]

Video chat has emerged as an important application on the Internet.  In this talk, we focus on the popular application of random video chat, where both online and mobile users are randomly matched.  We first introduce mvchat, an Android-based mobile client that we built to collect a large-scale data set from the popular Omegle random video chat service.  Then we discuss key characteristics of user behavior that we have discovered in random video chat, presenting the relationships between such factors as gender, camera use, audio type, users with long sessions, mobile accelerometer and sensor data, and interestingly normal vs flashing behavior.   We show how traditional classifiers designed to detect misbehavior in online random video chat perform poorly in a mobile environment.  We demonstrate that fusing mobile sensor data with multi-dimensional image data can significantly improve classifier performance, and illustrate the tradeoff between speed and accuracy in such a cascaded classifier.

Mais informações


retornar