CULTESC Research · Metric note
Played Past the End
Reading the extra seconds in short-form video.
A short video starts again. Maybe the joke deserves another look. Maybe a detail was missed. Maybe the viewer never noticed the restart.
At CULTESC, those possibilities lead to different edits. An ending that invites another look and an ending that leaves someone confused might both produce extra playback. The number alone cannot tell us which experience occurred.
Public datasets make the distinction concrete. KuaiRand records viewing time and video duration. KuaiRec records those fields and their ratio. Both originate from Kuaishou; they are external research sources, separate from CESC’s audience data.
A useful is recorded play time divided by video duration, greater than one. A hypothetical ten-second clip logging twelve seconds would qualify. That establishes time beyond one video length, rather than a deliberate rewatch or sustained attention.
Put duration beside the rate.
The threshold is relative to the clip. A few extra seconds represent a different fraction of a short video than of a long one. Before crediting a story or an ending, inspect the duration distribution and the definition of recorded play time.
Exposure belongs beside it too. KuaiRand includes randomly exposed videos and standard recommendation logs. KuaiRec’s small matrix was designed for near-complete user–item exposure. Those are different collection settings, rather than interchangeable samples of a current social feed.
Neither dataset, by itself, tells us how today’s Shorts, Reels, or TikTok audiences respond. A useful comparison needs its own population, reporting window, exclusions, and measurement definition.
Playback leaves the reason open.
Completion, extra play time, liking, profile entry, following, and later return describe different behaviors. A log containing several of them does not automatically establish their order or the effect of one on another.
A longer interaction also gives someone . To learn what they understood, remembered, or wanted to see again, collect evidence that measures those questions directly.
For a production team, that changes what gets revised. If the working question is clarity, compare comprehension. If it is recognition, test identification. If it is continued interest, observe a later choice to return. Playback can sit beside those measures, but cannot stand in for them. Improving the ratio without checking the experience risks optimizing a number whose meaning remains open.
The creative question starts where the count stops: what did the extra seconds mean?
An extra second can belong to a joke, a missed detail, or a restart that passed unnoticed. The production decision depends on knowing which question those seconds are being asked to answer.