Content-based filtering

Content-based filtering builds a picture of what someone likes from the properties of the things they have already watched: topic, language, format, length, the words in the title. It then looks for items whose properties line up. Collaborative filtering does the opposite job, grouping you with people whose behaviour overlaps yours and recommending what they watched. Large platforms blend both.

The useful part for a creator is what each approach needs before it can work. A content-based system can place a brand new upload immediately, because it can read the upload. A collaborative system has nothing to go on until enough people have watched, which is part of why the first hours of a video look so different from the weeks after.