Spotify recommender algorithms research blog

Overall this has been an interesting and inciteful look into how a user can navigate their way through Spotify's recommender system. We have examined how users can be pushed into a filter bubble through using Spotify's curated playlist and how the application will create personalised illusions that give the impresion that a user is discovering 'new music'.  I would say my experience with using Spotify's curated playlists was, at times, repetative and lacking variety. I noticed a narrowing effect within the first week of using the application and found that Spotify favours recommending new users contempary mainstream music of today. Later research would prove that Spotify's RA collects data from rising trends and users behaviour on and off the app. (Spotify, No Date) Alongside this, Spotify also state that musicians can pay for their songs to be recommended to targeted users. Stassen offers additonal research proving that artists record labels are influencing per...

My work has payed off and it seems that there are ways to bring new recommendations into your base recommendation pool. Granted this took a lot of time and meant I had to listen to the Happy Hour playlist on repeat. However, we now have three new playlists that include this type of music. This includes: Daylist (Funk Soul Groovy Evening) Daily Mix 2 Martha Reeves and The Vandellas Mix Along with this, It seems the throwback subheading, on the homepage, is experimenting with older music. As I now have a 'Born in The 90s' playlist. Previously, this subheading only contained music that went as far back as the 2000s. This does show, that the diversification tools, that Spotify include in their recommender systems, are able to expand a users music tastes, when given enough historical data. If this experiment was to happen for a longer period, and a bot was used instead of myself, then the results would perhaps show that: 'While Spotify's RA does seem to favour pushing users ...

After listening to the entirety of the Happy hour playlist, even letting the RA take control once it finished, it still seems that my overall recommendations haven't changed. The application notes my top mixes to be: 2010s Mix (A mixture of Rap and alternative rock) Mac Miller Mix (Chilled out rap music) Chill Mix (A lot of R&B with inclusions of chilled rap) House Mix (House and D&B) Fred Again Mix (Pretty much the same thing) Upbeat Mix (This one has a mixture of Rap, House and R&B) (I noticed that with this one the music seemed a lot more artificially recommended) (Very popular artists and songs) R&B Mix (The bane of my existence) Pop Mix (Sabrina and etc) (pretty reminiscing of the start of the experiment) I checked both versions of my Day list but they had still not included any new music. This last portion of the experiment is really showing how RA's take a preference to new music.  This quote from Linda O, on Medium, reasons a potential cause for this phe...

I listened to the happy hour playlists once more today. It is quite a long playlists so listening to it multiple times a day may seem like a challenge. However, I will do what I must to ensure that the data I collect in these final days of this experiment is interesting and related to my research. As of now, the RA has shown no signs of adding these older songs to my: Daylist, DJ, or any other playlist. My base curated playlist still largely consist of RAP, RNB, and Alternative rock.  I did check out my daylist today. In the morning it seemed to be Chill RnB music, while during the evening it became a "Doof Doof House Friday" playlist. The RA must think I am some chilled out raver that goes out every weekend.

Today I listened to the Happy hour playlist again. In hopes that the RA would take note over the last days of the experiment and include this music into my base recommendations.  What I have noticed, is at the start of the experiment. If you listened to something more than once that would become apart of your recommendation pool. However, as the experiment went it seemed increasingly difficult to get genres of music to stick. Examples of  'new music'  like the Lo-fi-house playlist never seemed to be brought into my base recommendations. This goes the same for the any of the new genres I encountered in the discover weekly playlists. Why is this.  As the experiment has gone one less and less "new music" has been recommended to my account and new music I am recommended doesn't become a part of my base recommendation pool.  I will attempt to change this over the last days and just see how many times, and for how long, I will have to listen to the Happy hour playlist...

Listening the the Happy hour playlist I found yesterday has been the most interesting part of this experiment. Offering me a playlist filled with popular songs from the 60s and 70s was not something I expected when going into this. This playlist was filled with Soul music, early Rock and Roll and the contemporary music of that time.  This is the RA showcasing its diversification tools first-hand and shows that you can experience little pockets of "new music" occasionally. Overall I feel like this will just become a personalised illusion and my base recommendations will stay the same. However, the fact that I have been recommended this is still very cool. It shows that the RA can diversify the content you are recommended and that there is a, ill be it small, way to find new genres and music that is different from your main recommendations on the app.