Video Recommendation: Ranking and Re-Ranking
Explore how to design effective video recommendation ranking systems that balance multiple objectives such as watch time, user satisfaction, content freshness, and diversity. Understand multi-head architectures, Wide & Deep and DCN models, re-ranking strategies, and fairness considerations to create scalable and fair recommendation systems for production environments.
The retrieval stage returns a few hundred candidate videos using a two-tower ANN lookup optimized for high recall over a large item catalog. The ranking stage must solve a different problem: score and order those candidates more accurately against multiple competing business and user objectives. This is the fine-grained scoring layer of the recommendation funnel. The core interview question this lesson answers is simple to state but difficult to implement: how do you design a ranking system that balances watch time, user satisfaction, content freshness, and diversity without allowing one objective to dominate the ranking policy? Large-scale video rankers face this challenge, and the same design patterns apply across many large-scale recommendation systems. This lesson covers multi-objective ranking formulation, Wide & Deep, and DCN-style architectures used for ranking, reranking for diversity and policy compliance, and fairness as a core design constraint.
Multi-objective ranking formulation
A single-score ranker that predicts only one thing, say expected watch time, will inevitably overfit to that signal. Users end up in autoplay rabbit holes, consuming increasingly sensational content. Satisfaction drops. Creators' gaming watch time gets disproportionate exposure. The system needs to predict multiple signals and combine them.
In a multi-objective ranking setup, the model produces several prediction heads from a shared backbone. Each head targets a distinct user engagement or satisfaction signal.
P(click): The probability that a user will click on the video thumbnail, capturing immediate interest.
E[watch time | click]: The expected watch duration given that the user clicks, measuring depth of engagement.
P(like) and P(share): Probabilities of explicit positive feedback, serving as proxies for genuine satisfaction.
P(not-dislike): The probability that the user does not actively dislike the content, filtering out low-quality recommendations.
The final ranking score combines these predictions through