Engagement Optimization in Large-Scale Video Platforms: A Systems Review of Metrics, Ranking, Experimentation, Latency, and Feedback Loops

Large-scale video platforms increasingly depend on engagement optimization to determine what users see, how content is ranked, and how platform value is measured. However, engagement is difficult to optimize because platforms rely on imperfect proxy signals such as click-through rate, watch time, completion rate, likes, and session continuation rather than directly observed user satisfaction. This paper presents a systems-oriented review of video engagement optimization across five interconnected layers: engagement metrics, multi-stage ranking architectures, online experimentation, latency-sensitive serving, and personalization feedback loops. The review synthesizes foundational and recent work on large-scale recommendation systems, controlled experimentation, tail latency, short-video engagement metrics, and feed auditing. The paper argues that engagement should be understood as an end-to-end systems property rather than only a modeling objective. It further highlights that ranking quality, serving responsiveness, metric design, and user feedback dynamics jointly shape observed engagement outcomes. The review identifies open challenges in long-term user-value measurement, performance-aware evaluation, feedback-loop mitigation, and commerce-specific video usefulness.