Whole Page Optimization with Global Constraints

Abstract

The Amazon video homepage is the primary gateway for customers looking to explore the large collection of content, and finding something interesting to watch. Typically, the page is personalized for a customer, and consists of a series of widgets or carousels, with each widget containing multiple items (e.g., movies, TV shows etc). Ranking the widgets needs to maximize relevance, and maintain diversity, while simultaneously satisfying business constraints. We present the first unified framework for dealing with relevance, diversity, and business constraints simultaneously. Towards this end, we derive a novel primal-dual algorithm which incorporates local diversity constraints as well as global business constraints for whole page optimization. Through extensive offline experiments and an online A/B test, we show that our proposed method achieves significantly higher user engagement compared to existing methods, while also simultaneously satisfying business constraints. For instance, in an online A/B test, our framework improved key metrics such as customer streaming minutes by 0.77% and customer distinct streaming days by 0.32% over a state-of-the-art submodular diversity model.