Algorithm, Search & Distribution

Cold start

Also called: cold start problem

Recommendation systems run on history, so anything brand new gives them nothing to work with. That is the cold start. It covers the opening hours of an upload, the first weeks of a channel, and the viewer who installed the app this morning and gets broadly popular material until they watch enough for a pattern to form. YouTube judges each upload on how the viewers it reaches respond right now rather than on what the channel managed before, so every video passes through a version of this regardless of who published it.

A quiet first day gets read as a punishment. Most of the time it is the system still working out who to show the thing to.

More in Algorithm, Search & Distribution

All 118 Algorithm, Search & Distribution terms

Trying to put this into practice? The Chewbr Knowledge Bank walks the whole upload workflow, phase by phase.