2
$\begingroup$

I've written a script to use youtube-dl to collect all view-counts and ratings from all listed videos on a youtube channel: https://gist.github.com/philshem/ca0b8f964458c63ee4f0e68c0f700f39#file-get_stats-sh

To compare ratings is relatively simple, because they are already averaged. But I would like to compare views, which is not clear to me because the number of views is related to how recently the video was released.

I am looking for a metric that normalizes somehow the age of the video with the number of views. I could do something simple, such as view_count * days_since_release, but I don't expect that to be statistically meaningful.

Is there an example metric for comparing historical, cumulative data?

views x rating

https://twitter.com/philshem/status/1295725069268910081

$\endgroup$
1

1 Answer 1

2
$\begingroup$

A simple option would be to use the average number of views per day, this way the value is normalized and independent of how recent the video is. Of course averaging is simplistic, since the views don't follow a linear trend.

A more advanced option would be to model the evolution of views across time (for instance assuming a normal distribution), then estimate the parameters for every particular video and compare the predicted total number of views. This would be more complex due to the difficulty of estimating the parameters from an incomplete distribution.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.