Pop Viz: The Effects of Spotify’s Discover Weekly Playlist

April 11, 2016

If you are a fan of my blog, I’m going to bet you are also a fan of Peter Gilks’ blog and saw this excellent viz looking at his Last.fm/Spotify data. He does a great job laying out how to keep/get this data for yourself. And I liked his viz so much that I decided to rip it off entirely, down to the little pic of me wearing headphones:

Peter had some other cool visualizations analyzing his taste. He had some nice ones about genre, which I would totally do if I didn’t have 3,520 distinct artists to categorize. I did have some success using import.io to scrape Allmusic.com to get that information for my Festival Finder viz last year, so maybe if I have time I’ll take a crack at doing that. One thing I did finally learn a little more about is LOD calculations, which I used to find when the first/last date that I played an artist was. With that I was able to make this cool gantt chart of my Top 20 artists. Check out how one day of mourning David Bowie was enough to put him in my Top 20 for Q1.

I’ve been using Last.fm since 2006, so I have nearly a decade’s worth of music listening data to look through. I could probably create a whole blog just on my personal music listening habits, but I doubt that anyone would be all that interested in that besides me. However, I did start to scratch the surface of an interesting. About a year ago, Spotify started making these weekly curated playlists for their users called “Discover Weekly.” Spotify uses all kinds of sophisticated recommendation engines to determine what to add to each user’s individualized playlists. After giving them a try for a while, I learned that Spotify’s robots KNOW THEIR SHIT. So, I made the Discover Weekly playlist a part of my music listening routine. As such, I’ve seen a bit of a jump in the number of new-to-me artists I listen to, especially on Mondays, when the playlist comes out. Check it out in the story below:

This is just scratching the surface of the kinds of analysis I want to do about Discover Weekly. Coming up, I’m going to see how often they get things right, if what I’m currently listening to has an effect on what Spotify recommends me, and how much of their playlists are actual new artists to me and not just tracks I don’t listen to as much from artists I already love.

#Main
2 Comments
Michael
April 18, 2016 @ 12:41 pm

I liked this a lot, thanks for making your workbook available to download! I’m still new to Tableau so personal projects help me learn stuff for work projects :) I scraped my last.fm data back to 2005 too, and was able to figure out some cool stuff. I wanted practice making a selector with string parameters so I did that between your First Listen Count and my own calc that just concatenates the Artist, Album and Song name for a unique string and then COUNTD that measure to determine Unique Track plays – dual axis that with SUM(Number of Records) and you effectively get the unique track play % for that time period.

I also made a second First Listen for just a single track using that unique track name concatenation to see how it changed compared to the First Listen Artists.

When I pop the First Listen Tracks or First Listen Artists running total alongside my Number of Records running total, the slope is almost a perfect match when axes are not synced, so at least I know I’m consistently listening to new stuff over time. Thanks Internet streaming!

Reply
April 27, 2016 @ 11:01 am

Awesome!! I tried getting all my last.fm data from them awhile back and it was painful. I’ve only got back to 2009, but still, that’s a lot of tunage! Especially since when I still used my iPod I was scrobbling from there, too. Ben’s data grab made that all easier. So I’m off to viz!

Reply

Leave a Reply to Alex Cancel reply

Your email address will not be published. Required fields are marked *


*