I recently got into a debate on how to best measure when my Facebook/Twitter/Instagram friends are most active. Imagine that I have the input data set that consists of (timestamp, user_id), where timestamp is the time when the user did something like posting a new status. Given this information, how would you measure when the most users would see your post? We can assume that we are going to quantize this to an hour or half hour interval.

My friend suggests that simply averaging the number of unique users engaged in any given hour isn't enough because not all users are active every day, but randomly during any given time of day, making all hours look the same in the long run. Thoughts?


closed as too broad by Xi'an, Michael Chernick, kjetil b halvorsen, John, mdewey Feb 27 '18 at 12:33

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • $\begingroup$ I assume that data scientists at Facebook can just query directly the number of views a post has. That's probably the best way to answer this question. Outside of that, I don't see any practical way to answer this question. $\endgroup$ – Sycorax Feb 26 '18 at 17:19
  • $\begingroup$ Are you saying that given this data, this question is impossible to answer? What additional data would you need? $\endgroup$ – ipartola Feb 26 '18 at 17:52
  • $\begingroup$ Social media companies don't serve content in time-series order, they serve it according to user interests and interactions. "Time of day" is only relevant if you also know the locations (i.e. local times) of potential consumers. Views are plausibly strongly correlated with follower/friend counts (because who else would see it?) etc etc $\endgroup$ – Sycorax Feb 26 '18 at 18:26
  • $\begingroup$ I mean hypothetically speaking, if we had the perfect time series dataset, and timezones where known. $\endgroup$ – ipartola Feb 26 '18 at 18:29

The mean number of active users per time interval is indeed a simple but sensible way to approach this problem with this kind of data. Fancier approaches could involve modeling user activity as a Poisson process.

Your friend's suggestion boils down to the idea that people are overall no more likely to be active on a website at one time of day than another. This is unlikely, as you can tell from publicly available website usage data, and would be easy to check if you actually had the data you wanted to analyze.


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