Friday, April 19, 2013

The Nuit Blanche Viewership Dataset

It's Friday, here is a homegrown dataset. I noticed the other day that I had a thousand days worth of data featuring the direct visits to the site as well as RSS feed views of Nuit Blanche. The data was gathered in some excel spreadsheet format they extend from July 7th, 2010 till April 18th, 2013:
The data includes the following: Day, Date, Nuit Blanche Page Loads, Nuit Blanche Unique Visits, Nuit Blanche First Time Visits, Nuit Blanche Returning Visits,  RSS Subscribers, RSS Reach, RSS Item Views,  RSS Item Clickthroughs and RSS Hits. The dataset has been added to the

Datasets and Challenges page.

I played with the MOUSSE package that was featured earlier this week and this is what I got in terms of number of leaves (subspaces) where the data live over time. The first graph shows the number of leaves when the data is not normalized where the second graph shows the number of leavers when the data is normalized. In both case, we see some increased complexity as we go along.

The normalization does skew things a bit as the blog was recipient to some friendly DDOSs', sudden changes to zero of viewership for one or two days, etc. In all, I think it somehow describes more the quality of the service of the platform than much of the variability in the actual statistics of the blog (which are always going up). Certainly, a more serious look at the data will provide some additional insight. 

Constants used for this run (using a version of runM.m) were:

numTrain = 10
epsilon = 5e-7;
lambda = 0.9;

No comments: