'Smart shuffle' should probably include more less popular songs (at least for certain listeners)

[allecco]

When I turn on smart shuffle, I am always wondering why the proposed music sounds so generic and not really fitting to my taste. Because of that, I checked how many plays the recommended songs typically have - the plays were between 250000 and multiple millions. I think the case is clear. Please, Spofiy, also propose music to me that does not have a high play amount but that fits my taste more. Thanks, and have a good one!

[Kincaed]

Plan

Premium

Country

United States

Device

PC (MS-7D77), Samsung S21

Operating System

Windows 11, Android

My Question or Issue

Last topic got moved to a clearly-unrelated thread (category and problem were different from mine), so I did my own troubleshooting. My Liked Songs playlist has 2311 songs from a variety of genres and artists, varying in popularity. Smart Shuffle for this playlist has historically been pretty good at matching this variation of many kinds (genre, popularity, time period, etc.) When I have a single Queens of the Stone Age song liked, my Smart Shuffle exclusively gives me Rock that charted in the 1980s-2010s- Black Sabbath, Cage the Elephant, Rage Against the Machine, Joy Division, Beck, etc. Most of the recommendations are unconnected to music I choose to listen to except that, broadly speaking, a lot of both are technically rock (though rarely the same sort). I have a pretty huge variety of genres, and I mean it when I say it adds, literally, only radio Rock that hit the charts. Unless I remove that single QotSA song. If I remove songs from other popular artists, the variation of recommendations improves even more. This represents a tiny handful of songs out of the total, but they clearly have a hugely disproportional impact on the algorithm's interpretation of my taste. Deleting a couple popular songs means that Smart Shuffle is magnitudes more likely to recommend music I like, from both artists I actually listen to, and unfamiliar ones like them. That is, the magnitude goes up at all from near-0. This demonstrates that something about the system, at least in my case, is very much not working.