Exploring Radio Paradise playlist data

I’ve been a listener and supporter of the internet radio station Radio Paradise for many years. I like the music of mostly rock mixed with other genres, including jazz and classical, selected by a professional DJ. And perhaps more importantly, I like that it’s free of ads with no-hassle optional listener support.

Playlist data

A few years ago, I started extracting the playlist data from their website. Why, I’m not sure. I imagined a few overly optimistic projects that haven’t come to pass.

  • computing relatedness of songs by how often they were played near each other
  • detecting the set breaks when the DJ says, “and now for something completely different”
  • looking for frequently paired songs
  • measuring the balance of the music mix against other radio stations (but I never found any other radio station playlist data)

Instead, I’ve shared some basic analyses in the past of top rated songs, ratings distribution, artists play trends, and play count versus rating.

Getting the data has become an end-of-year ritual for me that’s only possible because they keep their old website running at legacy.radioparadise.com. Like most radio stations with online playlists, the current site’s list is a dynamic page only showing the latest songs played. Their legacy site allows you to see the playlist at any time in the past using URL query fields.

Shift detection

While I haven’t been listening as much lately, I’m aware that Radio Paradise refreshed their format in 2022, and I’m wondering if I can detect the format change from the playlist data. Fortunately, the captured data includes the release year of each song played, so if the refresh changes the newness of songs played that should be detectable. As a first look, here’s a scatter plot of all 2 million song plays since 2007 and their release years.

Scatter plot of release year by song play date for 2 million songs. Beyond 1950 or so, it's just a solid mass.

That’s not so useful for detecting the programming change because of all the overstriking for the main release years 1960 – present, but it does show:

  • the classical music songs “released” in the 1700s and 1800s.
  • a fair number of early 1900s releases
  • a big gap of missing data in the data 2010 playlist
  • the maximum release year increases with the play year, as expected, except:

We can zoom in a bit by clipping release year to 1960, filtering out the earlier playlist years with data gaps and suspicious release year data. and using a heat map instead of the overstriking dots.

Heat map of yearly play counts from 2011 to 2022.
  • A few horizontal bands suggest that some years are more popular song sources, particularly 2002 and 1969-1972.
  • In the early years, 2011 – 2015, they were playing more new songs.
  • 2022 does show a renewal of playing more new songs, after a noticeable lull in the preceding years.

Zooming in a little more we can break it down by month for the playtime and see which month the new programming kicked in (adding a gray background and gridlines so the top cells and year boundaries can be seen):

Heat map of monthly play counts from 2018 to 2022.

Now we can see it was May of 2022 when the refresh happened, and it was stronger than apparent in the yearly view. It’s especially impressive given that in May 2022, there were only a few months of songs to choose from since 2022 was still in progress. About 10,000 song plays happen in each month, so 800 is about 1 in 12 songs, which doesn’t seem like a disruptive shift but it’s certainly noticeable.

I didn’t normalize the monthly counts for month length, so February has visibly fewer plays each year as seen by the lighter vertical bands.

I was only looking for the 2022 shift, but after seeing the difference in the early 2010s, I made this look at the whole time period, after computing for each day the percent of songs that were released within the past three years.

Scatter plot over 2007-2022 showing a downward trend from 30% to 5% then an uptick in 2022 to 10%.

So, though the 2022 refresh seems like a big change from recent years, it might be seen as a return to the prior normal. 2022 looks much lower than the earlier years, but that may be because its focus has been on only one of the previous three years.

Leave a Reply