Welcome to Dijemeric Visualizations

Where photography and mathematics intersect with some photography, some math, some math of photography, and an occasional tutorial.

Monday, July 28, 2014

The Rise and Fall of a Flickr Photo

If you post photos to Flickr, you can see the views accumulate and then dramatically decline until there are few or no daily views.  This is expected, but what determines how many views are received and how rapidly the decline occurs?  Is there a relationship between the views on an individual image and its associated album?  Is there a reverse relationship such that a popular image can bring views to other associated images in the album?

One of my images, On a Very Warm Day, got some extra attention when it was invited to the group Explore.  I hadn’t heard of it before, but once there the counts climbed dramatically, exceeding 1000 in a matter of hours.  I have a few images that exceed 1000 views, but it is usually after several months or even years.  While it is an appealing image, it was still a surprise.  But given this opportunity, I watched the statistics a bit more closely.

 On a Very Warm Day I tracked the views over the first day during the count climb and then over the next few days as the views declined.  It is obvious that initial fame is fleeting and the counts go down rapidly.  After nine days the daily views declined to under 50/day from the initial 2000+ on day 1.

Flickr Image Stats

When the Warm Day image views are plotted together with the views of all my Flickr posts, the two curves parallel each other (Chart I).

 Chart I: Image and Total Views Over a Nine-Day Interval

To see if there was a correlation, I plotted the cumulative image views against the cumulative total views (Chart II).  The correlation is 0.95 indicating a strong relationship between total views and image views.

 Chart II: Cumulative Image Views as a Function of Total Views

The fit indicates that for each additional 1000 total views, images views would increase on average by 444; at least during this 24 hour period.  This made me wonder if this relationship held over each of the time intervals for July 20.  For example, was it possible that during some of the time intervals, all of the views were due just to this one image?

 Chart III: Relative Increase in Image Views Based on Total Views Over Time
 Chart III shows the slope of the image views plotted against time where the slope is relative to the total views.  The maximum value at 1:11 pm was 1.3, which means for an increase in total views of 100 the single image view increase was 133 – an impossibility since the total views include the image views.  While the cumulative total views always exceeded the cumulative single image views, it should also hold that the increase in total views during any given time interval should never be less than the single image views.  So what now?  I suspect the discrepancy in the image view increase at hour 1:11pm is a bug in the Flickr views.  Initially I thought it was a typo on my part but have checked the numbers and don't think that is the case.  It makes sense that a strong image can bring in more traffic to other associated images, but this analysis makes a weak case to support that contention.   Nonetheless, I had fun doing the analysis and may try to determine the source of the mysterious discrepancy.