How machine learning powers Facebook’s News Feed ranking algorithm

This post details the machine learning architecture behind Facebook's News Feed ranking algorithm. It explains how multitask neural networks, embeddings, and offline learning systems are used to predict user engagement (likes, comments, shares) for billions of users in real-time. The system employs a multi-pass approach for scoring and ranking posts, including an initial lightweight model to select relevant candidates, a main scoring pass using powerful neural networks, and a contextual pass for diversity. The post also discusses how user surveys are used to align the ranking objective function with perceived long-term value.