Designing Video Recommendation Systems with Multi-Objective Ranking

Building a recommendation engine that only optimizes for clicks leads to clickbait and poor user retention. Modern video platforms require sophisticated ranking systems that balance competing goals like watch time, user satisfaction, freshness, and algorithmic fairness. This course guides you through the core engineering principles needed to solve these complex trade-offs. In this text-based course, you will learn the foundational concepts and architectural patterns required to design multi-objective ranking systems. You will understand how to transition from simple single-metric models to robust multi-task learning architectures that power today's leading streaming services. What you'll learn: - Understand the core phases of recommendation pipelines, focusing on the retrieval and ranking stages. - Formulate multi-objective optimization problems to balance watch time, explicit feedback, and long-term user retention. - Explore modern model architectures, including Multi-gate Mixture-of-Experts (MMoE), for multi-task learning. - Apply techniques to mitigate system biases, such as position bias, to ensure recommendation fairness and diversity. - Evaluate ranking model performance using both offline metrics and online evaluation methodologies. The course begins with foundational definitions of recommendation funnels and ranking objectives. You will then progress through detailed written explanations of multi-task network architectures, loss function design, and bias mitigation strategies. This course is designed for software engineers, data analysts, and aspiring machine learning practitioners who want to understand the mechanics of recommendation systems. Basic programming familiarity is helpful, but no prior experience with recommendation architectures is required. Start reading today to master the design patterns behind modern video recommendation engines.