Facebook recommendations stop guessing from clicks alone: Your algorithm lets people rewrite Feed and Reels in a sentence
Synopsis
Meta introduced Your algorithm, which lets Facebook users state in natural language what they want and do not want to see, with those preferences applied to both Feed and Reels. Previously the recommendation systems learned mainly from implicit engagement signals, which scale well but adapt slowly and capture little nuance. The system uses a collection of large language models to infer interests, validate preference changes and perform a final rerank, while a supplementary fetch and cache invalidation make changes take effect faster; Meta notes that interpreting ambiguous input and responding more precisely remain open problems.
Interpretation
Facebook recommendations can now be rewritten directly by users in natural language, rather than being determined only by implicit behavior such as clicks and dwell time. The earlier recommendation systems learned from implicit engagement signals, which scale well but adapt slowly and capture little nuance. The Your algorithm page splits interests into Interested In and Not Interested In lists, and every change applies to both Feed and Reels and updates further as the person gives feedback on individual posts.
A preference passes through inference, validation, storage and ranking before it affects what is shown, with different models handling each stage. In preference management, a smaller model checks that a phrase is a coherent topic, detects duplicates and maps it to a clean topic name, then enriches it with the embedding and search keywords ranking needs. A fine-tuned large language model reviews the ranked list and returns only the items to remove; Meta reports this small model was faster and more reliable than a general-purpose model judging each item individually.
Adding an interest and blocking an interest each have dedicated delivery mechanisms so a change does not wait for the next regular fetch. Before a new preference exists, content is not picked up until the next fetch request, and content already cached on the device was ranked before the preference was set. Adding to Interested In triggers a supplementary tail load outside the normal fetch cycle, and adding to Not Interested In makes the client obtain that interest's clusters and drop matching cached content before display.
Perspective
The system targets Facebook Feed and Reels recommendations, and the public description covers the mechanisms of interest inference, preference management, ranking and delivery, plus the division of labor among large language models; no effect metrics or online comparison data are given.
Meta says important opportunities remain, including better interpreting ambiguous input and making recommendations respond even more precisely to each request, and it is exploring additional entry points for shaping preferences.