A comprehensive data-driven approach to evaluating quality of experience on large-scale internet video service

Internet video is currently one of the most popular Internet services, accounting for more than 50% of the overall Internet traffic. Various access patterns of users worldwide contribute to increasing complex scenarios of Internet video. Under the circumstances, there emerges a valuable yet challenging topic that industry and academia keep focusing on: how to evaluate quality of experience (QoE) of Internet video in the wild? In this paper, we propose a comprehensive data-driven approach to evaluate QoE based on massive dataset extracted from a large-scale Internet video service provider. With the methods of feature engineering and fuzzy theory, we exploit user's situational features and session's quality features. Then we conceive an appropriate QoE evaluating model called bagging-based Bayesian factorization machine to correlate the aforementioned features with user's QoE. The experimental results demonstrate that our approach is adaptive for QoE evaluation on Internet video both in efficiency and effectiveness. Moreover, our approach achieves higher degree of accuracy compared with baseline methods, including what associated works present.