Your Algorithm: Giving People More Ways to Personalize Facebook Recommendations

Why it matters

Facebook has published a technical account of "Your algorithm," a suite of AI-powered tools that lets people shape their Feed and Reels recommendations by stating preferences in natural language. The system departs from the traditional approach of learning only from implicit engagement signals, which adapts slowly and captures little nuance. A collection of large language models understands, stores, and tracks stated preferences alongside user behavior. The interface shows a summary of recent viewing, a list of inferred interests, and a list of rejected topics, with every change applied responsively to ranking and delivery. The engineering challenges are notable: new interests must surface relevant content quickly without dominating the feed, achieved through dedicated retrieval sources, relevance checks, and a control layer that spaces out and caps each interest's contribution. Exclusions are enforced through layered filtering across ranking, reranking, and the device, with special handling for quality-related requests and named creators. Supplementary fetches and cache invalidation ensure changes take effect promptly. The piece matters because it details how a major platform is making recommendation systems more transparent and directly steerable.