How Recommendation Feeds Choose Videos

A video feed can find candidates in two very different ways: by following viewers who watched what Maya watched, or by matching videos with similar learned representations. Neither decides the final order.

Maya’s viewing history resembles these peers’ histories. Because they watched the pasta video, collaborative filtering can offer it as a candidate for Maya. Shared behavior is a useful signal, not proof she’ll want it.

An embedding is a learned numerical representation of a video. Here, quick-dinner and one-pan rice sit close together, while the guitar video is farther away. Maya’s watch can therefore retrieve rice as a candidate.

Maya's candidate pool gets two plausible videos through different routes: peer behavior points to Pasta, while video similarity points to Rice. Candidate retrieval narrows a large catalog to items worth evaluating; it doesn't decide which one should appear first.

Let's think this through. Maya watched a quick-dinner video. Which method can suggest a nearby one-pan rice video without relying on similar viewers? A: Collaborative filtering. B: Embedding similarity. C: Final ranking. Choose an answer, or just think it through. I'll explain in a moment.

The answer is B: Embedding similarity. Embeddings let the system retrieve videos with nearby learned representations. Collaborative filtering uses patterns from other viewers, while ranking orders candidates after they have been found.

Pasta and rice enter Candidates through different signals: similar viewers and video similarity. The Ranker then evaluates each using Maya’s recent pasta save alongside other available signals, instead of simply preserving the order in which they were retrieved.

In this illustrative feed, the ranker puts pasta ahead of rice, using Maya’s recent pasta save as one signal. That order isn’t guaranteed; other signals and serving rules can change what appears first.

Find candidates, then order them. Collaborative filtering suggests pasta, embeddings suggest rice, and the ranking puts pasta first in Maya’s feed.