How recommendations are made
Most services combine several of these and decide both which rows to show and the order of titles within each row.
- Collaborative filtering: viewers who watched what you watched also watched this
- Content-based: this title shares a genre, cast, mood or theme with ones you finished
- Popularity and trending: what many people are watching now
- Editorial: collections chosen by people
- Contextual: adjusted for device, time of day and profile
The cold-start problem
A new viewer has no history, and a new title has no audience yet. For new viewers, start from popular and editorial choices, or ask for a few preferences at sign-up. For new titles, rely on metadata until viewing data builds up. A small or new service leans more on content-based and editorial methods for the same reason.
Why it matters and how to judge it
A viewer who cannot find a second thing to watch leaves. Good recommendations shorten the time between opening the app and pressing play, and spread viewing beyond the few titles on the front page.
Judge it by outcomes: the share of plays that began from a recommended row, time to first play, and whether viewers shown recommendations watch more and stay longer.
Last reviewed