Exploiting Danmu Interactions for Optimizing Crowdsourced Livecast Services

The past decade has witnessed dramatic growth of crowdsourced live broadcasting (livecast) service across the world. Many livecast platforms provide synchronized interaction mechanism for promoting viewer engagement. In such mechanism, viewers are quite heterogeneous in terms of watching channel preference and sensitivity on quality of experience (QoE), and their perceived service qualities can be easily affected by interactive message density and feedbacks from others. In this paper, we build a new quality model to capture these unique factors, and propose a two-level decision making strategy to cost-effectively map viewers to suitable edges. We cluster viewers based on their QoE sensitivities, and schedule for clusters rather than for individual viewers. Trace-driven experiments show that our method has comparable performance to individual based scheduling method, and is superior to other baselines.