Multi‑objective Optimization for Guaranteed Delivery in a Video Service Platform

ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) is a premier international academic conference in data mining, classified as a CCF A‑class conference. Founded in 1995, it is held annually. The 2020 edition took place from August 23 to 27 in San Diego, California, USA. The Research Track received 1,279 valid submissions and accepted 216 papers, resulting in an acceptance rate of approximately 16.9%.

A paper from Alibaba Entertainment’s foundational platform, titled “Multi‑objective Optimization for Guaranteed Delivery in a Video Service Platform,” was accepted to the KDD 2020 Research Track. The study addresses the complex, nonlinear, and chaotic nature of content exposure and click behavior, which depend not only on content quality but also on update timing, update strategies, and user click habits. By analyzing the exposure saturation effect, the authors introduce a novel hot‑content exposure sensitivity model based on differential equations. Building on this model, they propose a multi‑objective optimization framework and algorithm that operate under exposure‑resource constraints, ensuring guaranteed delivery of video content. This solution has already been deployed in multiple scenarios within the Youku platform.

Authors: Hang Lei, Yin Zhao, Long‑Jun Cai.