Â鶹¹ÙÍøÊ×Ò³Èë¿Ú

Last Friday the Â鶹¹ÙÍøÊ×Ò³Èë¿Ú's RAD (Rapid Application Development) Unit, led by George Wright, held two days of hacking in a Recommendation Super Sprint - the aim being to get people from around the Â鶹¹ÙÍøÊ×Ò³Èë¿Ú to play around with recommendations and personalisation. Chris and I from the R&D team at Radio Labs went along for the first day with few concrete ideas but the desire to play with some of the data we've gathered from .

Chris took a dump of the Radio Pop database - we've got around 1400 registered users and 24,000 "listen events", from when we launched in September to now. He removed any personal and extraneous data and then used PHP to process the data and write out text files of the data we needed - basically, a piece of data for every hour in which each user listened to Â鶹¹ÙÍøÊ×Ò³Èë¿Ú Radio through Radio Pop. I then used to draw some images of this data.

Click to see the full-size image, it's much more interesting

The first image shows everyone's listening. Each row of pixels represents an individual user of Radio Pop, with time going from left to right. Each pixel then represents one hour of one day, plotted in the appropriate position - it is black if they weren't listening and it is coloured in if the user listened during that hour, the colour representing the radio network they tuned to (e.g. Radio 4 is dark blue, Radio 1 is white; see the key at the bottom of the image).

The distinctive curved line is because the users (i.e. rows of pixels) are in order of registration on the website, so those at the bottom joined Radio Pop later. And people tend to listen a few times once they've registered, whether they continue to use it or not.

Download the full-res version and zoom in to see more detail, maybe see if you can find yourself. There are some interesting patterns of recurring listening and combinations of networks and patterns are one of the reasons for visualising data as the human brain is tuned to pick out regular recurrences in images. Here are some things I noticed...

These groups of 5 white dashes seem to show someone listening to a few hours of Radio 1 every week day - The Chris Moyles Show I guess?


Someone listening every day in batches, with a selection of networks.

Again, click to see the full-size image

This second image is a modification of this where we removed the dependance on time so each row is now all the listening for that user represented as pixels stacked alongside each other. It makes it a lot easier to see how much people have listened and to what. This produces even more pretty patterns with the coloured stripes showing regular listening patterns...

My listening is the second row from the top of the image, I normally listen to Radio 4.

Here's someone with a definite regular pattern of listening, mainly Radio 4 with regular Five Live and 6Music outings.


I found this one interesting, it seems to show several days worth of listening to just Radio 3, then several days listening to Radio 2 and then 1Xtra.

Then finally we just plotted a timeline of the listening for each radio network, with a fuzzy blob representing how many people were listening at that time. Again the peak is just when we launched. Next? More data please!

A couple of caveats: 1) The data is only from users of Radio Pop and it isn't representative of Â鶹¹ÙÍøÊ×Ò³Èë¿Ú Radio listening. 2) None of the data is completely accurate, for instance by having a resolution of one hour per pixel we had to fudge what was listened to in an hour. 3) We haven't taken account of when on-demand listening actually happened, we use the original broadcast time of the programme.