On the effectiveness of video prefetching relying on recommender systems for mobile devices

Abstract

High quality video streaming over cellular networks is expensive for users and often limited by data caps. Offloading the traffic over other cheap wireless networks, e.g. WLAN, would be beneficial however they are often not available for mobile streaming sessions. A potential solution is to load video to a mobile device before the user accesses it, which is called prefetching. Prefetching requires indicators to perform a prediction what a single user is going to watch in the future. This prediction determines which videos are loaded to the users' mobile device. We leverage recommender systems of the video sharing site YouTube to design and evaluate mobile prefetching algorithms. A user study has been performed over a month in which users' YouTube accesses were supported by our prefetching algorithms which allowed for significant reductions in data usage on mobile networks while having a low computational overhead.