Incoming!

It started by playing with Processing to create some visualisations. BBC Radio get quite a lot of incoming SMS text messages and the excellent BBC R&Mi software architecture team have ensured we have really useful feeds of the incoming texts. Put the two together and you get a constantly changing view of how the audience is interacting with the radio.



This is a timelapse video of the most common words found in the text messages sent to BBC Radio 1 from lunchtime on the 19th July to the following morning. As it's less than seven days ago you could even relive the whole experience with the RadioPlayer! This is the schedule for the time that the video covers:

13:30 (when it starts) Colin & Edith
16:00 Scott Mills show (JK & Joel sitting in)
inc. 17:45 Newsbeat
19:00 Zane Lowe
21:00 Bobby Friction & Nihal
23:00 One Music with Ras Kwame
01:00 Soul Nation with Trevor Nelson
03:00 Radio 1 documentary
04:00 JK & Joel
07:00 Chris Moyles (Scott Mills sitting in)

The DJ names are fairly prominent and you can see lots of comments about the hot summer weather and how to keep cool. It then goes pretty quiet overnight, with some bursts of activity, before coming back to life around 7 in the morning. I like the way it gives an insight into how the audience are reacting to a show. It's also very unpredicatable - sometimes there's no activity and then the DJ will ask people to text in and all hell breaks loose. Watching it evolve in real-time is surprisingly compelling though unfortunately I can't provide this publically - partly technical and partly privacy (you'd be surprised what people put in their text messages to Radio 1). I'm looking at ways to safely represent this on the web (e.g. on /radio1/) while ensuring that it doesn't lose its compelling live nature.

Incoming messages appear on a spread message bus and are filtered to remove any stop words (e.g. "the", "at" or "a" plus the sms- and radio- specific "plz", "play" etc.) and then a running total is kept of the frequency of all the remaining words. The top 15 words at any one time are drawn with a random position and a size proportional to the word frequency. Then every 5 minutes the frequency for each word is reduced, and any that aren't common enough are removed, so that the words seem to decay. This means that the same word coming in regularly won't keep getting bigger but should stay constantly prominent. There are loads of tweaks that I could make to this algorithm, it's by no means perfect. I left it running and captured a screendump every 5 minutes (using Processing's saveFrame() command) and then turned the captures into a video with 2 frames per second - so that's 6 seconds of video for every hour of radio programming.

It's also worth noting how (relatively) easy it was to build this:

1. The R&Mi Architecture team make it easy to use incoming texts
2. Spread makes it easy to subscribe and receive messages
3. Processing (and Java) makes it easy to visualise things
4. Windows Movie Maker (!) makes it easy to create a movie from a sequence of still images
5. YouTube makes it easy to put it on the web
6. Blogger makes it easy to write about it